Image transmission method and system for digital camera

Abstract

The invention relates to the technical field of digital televisions, in particular to an image transmission method and system for a digital camera. The method comprises the following steps: acquiring original image data shot by a digital camera; performing target scanning processing on the original image data to obtain target feature data containing target object type and position information; performing weight calculation processing on each target object according to the target feature data to obtain target importance data; thermal diffusion processing with the center point of the target object as the reference is conducted on the original image data according to the target importance data in a Gaussian distribution mode, importance thermodynamic diagram data are obtained, and the thermal diffusion processing uses a maximum value superposition mode when thermal distribution of the multiple target objects is overlapped. According to the generated thermodynamic diagram data, a differential compression method is adopted, so that the problem that image details are lost under the condition that the bandwidth is limited in a traditional single compression algorithm is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of digital television, and in particular to an image transmission method and system for a digital camera. Background Art

[0002] A digital camera is a camera that uses electronic sensors to convert optical images into electronic data. Image transmission refers to the process of transferring image data from one device to another, and is widely used in digital cameras, monitoring systems, medical imaging and other fields. In digital cameras, image transmission usually involves the following key steps: First, the image is collected by the camera's sensor and converted into a digital signal, and then compressed and encoded (such as JPEG or HEIF format) to reduce the amount of data for easy transmission. Next, depending on the different transmission methods, the image data can be sent to the target device via a wired connection (such as USB, HDMI) or wirelessly (such as Wi-Fi, Bluetooth, NFC). The advantages of wired transmission are fast and stable speed, which is suitable for the transmission of large amounts of data; while wireless transmission is more flexible and convenient, suitable for instant sharing and transmission between mobile devices. In addition, modern digital cameras also support uploading images to the cloud through cloud storage services, and users can access and download these images on any device that supports network connection. There are many ways to transmit images in digital cameras, and different transmission methods can be selected according to different usage scenarios and needs.

[0003] However, in news reporting and sports event broadcasts, photographers need to quickly transmit photos taken on site to the news editing department. The biggest technical challenge in this scenario is the uncertainty of the network environment. Outdoors or at large-scale events, network signals are often unstable, which requires the transmission solution to have strong fault tolerance and automatic retransmission mechanisms, while also ensuring basic image quality under limited bandwidth. Summary of the invention

[0004] Based on this, it is necessary for the present invention to provide an image transmission method and system for a digital camera to solve at least one of the above technical problems.

[0005] To achieve the above object, an image transmission method for a digital camera comprises the following steps:

[0006] Step S1: acquiring original image data taken by a digital camera; performing target scanning processing on the original image data to obtain target feature data including target object type and position information;

[0007] Step S2: performing weight calculation processing on each target object according to the target feature data to obtain target importance data, wherein the weight calculation processing specifically includes determining the weight value based on the expression feature of the person when the target object type is a person, determining the weight value based on the motion trajectory when the target object type is a moving target, and determining the weight value based on a preset rule when the target object type is text information;

[0008] Step S3: using Gaussian distribution method to perform thermal diffusion processing on the original image data based on the target importance data with the center point of the target object as the reference to obtain importance heat map data, wherein the thermal diffusion processing uses a maximum value superposition method when the thermal distributions of multiple target objects overlap;

[0009] Step S4: Performing layered compression processing on the original image data according to the importance heat map data to obtain layered compressed image data, wherein the layered compression processing is specifically performing lossless compression on areas where the thermal value is greater than a preset first threshold, performing medium compression ratio compression on areas where the thermal value is between the first threshold and a preset second threshold, and performing high compression ratio compression on areas where the thermal value is less than the second threshold.

[0010] The present invention can accurately identify the target object in the image and its type and position information by acquiring the original image data and performing target scanning processing. This process lays the foundation for subsequent processing, so that the system can analyze and process the key content in the image in a targeted manner. By marking the type and position of the target object, the system can better understand the semantic information of the image, thereby providing a basis for subsequent weight calculation and compression strategy. This preliminary analysis of the image content enables the subsequent processing to focus more accurately on important visual elements, avoids the invalid processing of irrelevant information, and improves the efficiency of the entire system. In the target weight calculation process, different weight determination methods are used according to the types of different target objects, further highlighting the importance of key information in the image. For human targets, by analyzing the facial features of the human to determine the weight value, the emotional state of the human in the image can be captured, thereby focusing attention on areas with more emotional value. For example, a smile or a surprised expression often attracts the audience's attention, so a higher weight is given to ensure that these areas are better retained in subsequent processing. For moving targets, the weight value is determined based on its motion trajectory, and the dynamic elements in the image can be identified, which is crucial for capturing wonderful moments and recording sports scenes. The analysis of motion trajectories enables the system to identify targets that move quickly or have special motion states, such as acceleration or turning movements. These targets usually have high visual importance and are therefore given higher weights. For text information, the weight value is determined by preset rules to ensure that the text content in the image is given full attention. Text information often carries important semantic content, such as scores, time and other key data. By matching the rule library, the system can identify the importance of these texts and assign corresponding weights. This method of calculating weights according to different target object types enables the system to more intelligently identify important elements in the image, providing precise guidance for subsequent thermal diffusion and compression processing. Using Gaussian distribution for thermal diffusion processing can intuitively present the importance information of the target object in the form of a heat map. Thermal diffusion based on the center point of the target object can form an importance distribution area centered on the target. This distribution method conforms to the characteristics of human visual perception and can better simulate the degree of attention of the human eye to important areas of the image. When the thermal distributions of multiple target objects overlap, the maximum value superposition method is used to ensure that the importance of the overlapping area is fully reflected, avoiding the loss or underestimation of important information. The importance heat map data generated in this way provides an important reference for the subsequent layered image compression, allowing the system to judge the importance of the image area according to the heat value, thereby achieving a more accurate compression strategy. In the layered compression process, the original image data is layered according to the importance heat map data, which can achieve differentiated compression of areas of different importance.The area with a thermal value greater than the preset first threshold is regarded as a high-importance area, and lossless compression is performed on it to ensure that the image quality of these key areas is not lost, and the most important and visually valuable parts of the image are retained. The area with a thermal value between the first threshold and the second threshold is a medium-importance area, and a medium compression ratio is used to compress it, which can not only save a certain amount of storage space and transmission bandwidth, but also ensure the basic quality of the image and meet general visual needs. The area with a thermal value less than the second threshold is a low-importance area, and a high compression ratio is used to compress it, further reducing the occupation of storage and transmission resources. At the same time, since the importance of these areas is low, even after high compression, the visual effect of the overall image is less affected. This hierarchical compression strategy can minimize the redundancy of image data, improve the efficiency of image transmission, save storage space and transmission bandwidth, and is particularly suitable for the image transmission needs of digital cameras under different network environments. It can achieve high-quality image transmission under limited resources. In summary, this method realizes intelligent processing and optimized transmission of image data through a series of steps such as target scanning, weight calculation, thermal diffusion and layered compression. It can not only highlight the key information in the image, but also improve the transmission efficiency while ensuring the image quality. It meets the diversified needs of digital cameras in image transmission and demonstrates its advancement and practicality in the field of image processing.

[0011] The present invention also includes an image transmission system for a digital camera, which is used to execute the above-mentioned image transmission method for a digital camera. The image transmission system for a digital camera includes:

[0012] The image acquisition module is used to obtain the original image data taken by the digital camera; perform target scanning processing on the original image data to obtain target feature data including target object type and position information;

[0013] A target weight calculation module is used to perform weight calculation processing on each target object according to the target feature data to obtain target importance data, wherein the weight calculation processing specifically includes determining the weight value based on the facial expression features of the person when the target object type is a person, determining the weight value based on the motion trajectory when the target object type is a moving target, and determining the weight value based on a preset rule when the target object type is text information;

[0014] A heat map generation module is used to perform heat diffusion processing on the original image data based on the target importance data using a Gaussian distribution method with the center point of the target object as a reference to obtain importance heat map data, wherein the heat diffusion processing uses a maximum value superposition method when the heat distributions of multiple target objects overlap;

[0015] The image layered compression module is used to perform layered compression processing on the original image data according to the importance heat map data to obtain layered compressed image data, wherein the layered compression processing is specifically to perform lossless compression on the area where the thermal value is greater than a preset first threshold, to perform medium compression ratio compression on the area where the thermal value is between the first threshold and the preset second threshold, and to perform high compression ratio compression on the area where the thermal value is less than the second threshold.

[0016] The image acquisition module in the present invention is the starting point of the entire system. It is responsible for acquiring the original image data taken by the digital camera and performing target scanning processing on these data. This process can accurately identify the target object in the image and its type and location information, providing basic data for subsequent processing. Through target scanning, the system can understand the semantic content of the image, thereby providing a basis for subsequent weight calculation and compression strategy. This preliminary analysis of the image content enables the system to process the image more intelligently, avoids invalid operations on irrelevant information, and improves the overall processing efficiency. The target weight calculation module performs weight calculation on each target object according to the target feature data. This process can dynamically adjust its importance weight according to the type of the target object (such as a person, a moving target or text information). For human targets, the weight calculation based on expression features can capture emotionally rich areas; for moving targets, the weight calculation based on motion trajectories can highlight key moments in dynamic scenes; and for text information, the weight calculation based on preset rules can ensure that important semantic content is retained. This differentiated importance assessment enables the system to more accurately identify key information in the image, providing scientific guidance for subsequent thermal map generation and compression processing. The heat map generation module uses Gaussian distribution to perform heat diffusion processing on the target importance data to generate an importance heat map. This process can not only intuitively present the importance distribution of the target object, but also process the overlapping areas of the heat distribution by maximum value superposition to ensure that the importance information will not be underestimated. The generation of the heat map makes the key areas in the image clear at a glance, providing a clear reference basis for the hierarchical compression module, so that the compression strategy can focus on important information more accurately. The image hierarchical compression module performs hierarchical compression processing on the original image according to the importance heat map data. By setting different thresholds, the system can divide the image into high, medium and low importance areas, and use lossless compression, medium compression ratio compression and high compression ratio compression respectively. This hierarchical compression strategy can minimize the redundancy of image data and optimize storage and transmission efficiency while ensuring the integrity and high quality of key information. Lossless compression of high-importance areas ensures the complete retention of core information, while high compression ratio processing of low-importance areas further saves resources. This refined compression method significantly improves the overall performance of the system. Through the collaborative work of modules such as image acquisition, target weight calculation, heat map generation and image layered compression, this system can not only accurately identify and process key information in the image, but also achieve high-quality image transmission and storage under limited resources, significantly improving the intelligence level of digital camera image processing and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0018] Figure 1 A schematic diagram of the steps of the image transmission method for a digital camera of the present invention;

[0019] Figure 2 for Figure 1 Detailed step flow diagram of step S1;

[0020] Figure 3 for Figure 1 Detailed step flow chart of step S2 in FIG. DETAILED DESCRIPTION

[0021] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0022] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0023] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0024] To achieve this, please refer to Figures 1 to 3 The present invention provides an image transmission method for a digital camera, the method comprising the following steps:

[0025] Step S1: acquiring original image data taken by a digital camera; performing target scanning processing on the original image data to obtain target feature data including target object type and position information;

[0026] In this embodiment, firstly, the light signal in the scene is captured in real time by the built-in image sensor of the digital camera (for example, a CMOS sensor), and the captured analog signal is converted into a digital signal by an analog-to-digital converter to generate initial pixel data; then, the initial data are preprocessed, including according to a preset white balance (for example, 6500 Kelvin), color correction, noise reduction (3×3 median filtering can be used) and image sharpening processing to ensure that the image quality meets the subsequent target detection requirements; then, the preprocessed image data is scanned by a target detection algorithm based on a deep convolutional neural network (such as SSD or YOLO), the target object in the image is automatically identified, and the bounding box and position information of each target are extracted, and the identified targets are classified according to categories (such as people, moving targets, text information, etc.), and finally target feature data including the target object type and its specific coordinate position in the image is generated; in a specific application scenario, for a 1920×1080 high-definition image, these preprocessing parameters and detection algorithms can be adjusted to a real-time operation state to ensure that the detection results are accurate and timely.

[0027] Step S2: performing weight calculation processing on each target object according to the target feature data to obtain target importance data, wherein the weight calculation processing specifically includes determining the weight value based on the expression feature of the person when the target object type is a person, determining the weight value based on the motion trajectory when the target object type is a moving target, and determining the weight value based on a preset rule when the target object type is text information;

[0028] In this embodiment, according to the target feature data obtained in step S1, a weight calculation process is performed on each target object in the image, thereby generating target importance data; for human targets, the system extracts facial expression features through the embedded face detection and expression recognition module, and when a positive expression such as a smile or surprise is detected, the basic weight of the target (for example, a basic value of 1.0) is increased to 150%, otherwise it is kept as the basic weight; for moving targets, the continuous frame target tracking technology is used to calculate the motion trajectory, and by comparing the displacement and acceleration of the target in the continuous frames, when the target is detected to be in accelerated motion or there is an obvious turning action, its weight is adjusted to 130% of the basic weight, otherwise it is kept as the basic weight; and for text information targets, the system calls a preset rule library to compare the recognized text content, such as when score data or time information is detected, the target weight is set to 200% or 180% of the basic weight, respectively, and other text is set to 120% of the basic weight; the weight values ​​of all targets are linearly normalized to the range of 0 to 1 after calculation, providing a unified scale for subsequent processing.

[0029] Step S3: using Gaussian distribution method to perform thermal diffusion processing on the original image data based on the target importance data with the center point of the target object as the reference to obtain importance heat map data, wherein the thermal diffusion processing uses a maximum value superposition method when the thermal distributions of multiple target objects overlap;

[0030] In this embodiment, the Gaussian distribution method is used to perform thermal diffusion processing on the original image data according to the target importance data obtained in step S2, and the center point of each target object is used as the diffusion center; the specific operation is that for each target, the diffusion radius is first calculated according to its normalized weight value, and the radius size is determined by, for example, the diffusion radius increases by 10 pixels for every increase of 0.1 weight value, and then the corresponding two-dimensional Gaussian kernel is generated according to the radius, which means that when the diffusion radius increases by 20 pixels, the standard deviation of the Gaussian kernel increases by one unit to ensure the smoothness and attenuation characteristics of the diffusion template; then, the generated Gaussian diffusion template is convolved with the original image data with the target center as the base point to obtain the thermal distribution data of a single target; in the overlapping area of ​​multiple targets, the system compares the thermal values ​​of the convolution results of each target for each pixel point, and determines the final thermal value by the maximum value superposition method, avoiding the excessive enhancement that may be caused by simple superposition, thereby generating a complete thermal map data reflecting the importance of each target, and its parameters can be adjusted according to needs in specific applications, for example, the maximum diffusion radius can be set to 100 pixels to maintain the independence between different targets.

[0031] Step S4: Performing layered compression processing on the original image data according to the importance heat map data to obtain layered compressed image data, wherein the layered compression processing is specifically performing lossless compression on areas where the thermal value is greater than a preset first threshold, performing medium compression ratio compression on areas where the thermal value is between the first threshold and a preset second threshold, and performing high compression ratio compression on areas where the thermal value is less than the second threshold.

[0032] In this embodiment, according to the importance heat map data generated in step S3, the original image data is subjected to hierarchical compression processing to achieve a balance between data compression and image quality. Specifically, the original image is first divided into image blocks of equal size (for example, 16×16 pixels per block), the average heat value of each image block is calculated, and the image blocks are classified according to two preset thresholds: when the average heat value of the image block is higher than the first threshold (for example, 0.7), the area is regarded as a high-importance area, and lossless compression technology (such as PNG compression) is used to ensure that image details are not lost; when the heat value of the image block is between the first threshold and the second threshold (for example, 0.4 to 0.7), the medium compression ratio JP compression is used. EG compression, its compression quality parameter can be adjusted by linear mapping according to the thermal value (for example, the quality parameter is set at about 80); when the thermal value of the image block is lower than the second threshold (for example, lower than 0.4), the area is considered to be a low-importance area, and a high compression ratio is used. The compression ratio may be further improved by increasing the quantization step size (in an exponential growth relationship) (for example, the quality parameter is reduced to about 50); finally, the compressed data blocks are recombined in a preset order, and the compression type and parameter information of each block are recorded in the data header to generate the final layered compressed image data, thereby ensuring high-quality display in key areas while achieving a significant reduction in the overall data volume, which is suitable for real-time transmission scenarios with limited bandwidth.

[0033] Preferably, step S1 comprises the following steps:

[0034] Step S11: using the image sensor of the digital camera to collect the light signal of the current scene, and generating initial pixel data through analog-to-digital conversion;

[0035] In this embodiment, the digital camera uses a built-in CMOS image sensor to capture light signals in the current scene in a progressive scanning mode. The size of each pixel of the sensor is approximately 1.4 microns, ensuring high sensitivity and low noise performance; then, the collected analog photoelectric signal is converted by a built-in high-precision analog-to-digital converter. The analog-to-digital converter uses a 12-bit resolution to convert the analog voltage signal of each pixel into a digital signal to generate initial pixel data stored in a matrix form, where the value range of each pixel is 0 to 4095, and the exposure time is usually set to 1 / 60 second, so as to capture image information with sufficient details under normal lighting conditions.

[0036] Step S12: performing white balance adjustment processing on the initial pixel data to obtain white balance correction data;

[0037] In this embodiment, the system first analyzes the average brightness values ​​of the red, green, and blue color channels in the initial pixel data, and compares these average values ​​with the preset standard white reference value. Then, according to the comparison result, the corresponding linear gain adjustment is applied to each color channel. For example, when the indoor light is warm, the blue channel gain may be increased to 1.1 and the red channel gain may be slightly reduced to 0.95. This ratio-based adjustment method is used to compensate for the color temperature deviation caused by different light sources, thereby generating image data that has been corrected for white balance, ensuring that the overall color of the image tends to be natural.

[0038] Step S13: performing color correction processing on the white balance correction data to obtain color correction image data;

[0039] In this embodiment, the system uses a preset color conversion matrix to perform further color correction on the data that has completed white balance adjustment. The matrix is ​​carefully designed based on the characteristics of the camera sensor and the standard sRGB color space. By performing weighted linear combinations of the red, green, and blue components of each pixel (for example, the red component may be multiplied by 0.9, the green component may be multiplied by 1.0, and the blue component may be multiplied by 1.1), the color deviation caused by the uneven spectral response of the sensor is effectively eliminated, ensuring that the output image can accurately present the color of the real scene on different display devices, and finally generating color-corrected image data.

[0040] Step S14: performing image noise reduction processing and image sharpening processing according to the color-corrected image data, thereby obtaining original image data;

[0041] In this embodiment, the system first applies adaptive noise reduction processing to the color-corrected image data, adopts the bilateral filtering method and takes a filtering window of 5×5 pixels as an example, and sets the spatial domain standard deviation to 2.0 and the pixel value domain standard deviation to 25, so as to retain the image edge details while smoothing the noise; then, the system uses the unsharp mask technology to sharpen the denoised image, and by calculating the local contrast in the image, the edge detail information after appropriate attenuation is superimposed back to the original image, and the sharpening intensity is controlled between 0.8 and 1.2 to achieve a clear and detail-rich image effect, and finally obtain high-quality original image data for subsequent target detection.

[0042] Step S15: performing target recognition processing based on an artificial intelligence target detection network on the original image data, thereby obtaining target bounding box data;

[0043] In this embodiment, the system uses a pre-trained deep convolutional neural network target detection model (such as an architecture based on YOLOv4 or Faster R-CNN) to perform target detection processing on the original image data. First, the image is adjusted to the input size required by the network (for example, adjusted to 416×416 pixels), and then image features are extracted through multi-layer convolution and pooling operations. The bounding box position and confidence score of each target are simultaneously predicted at the detection head. Each target bounding box data includes the coordinates of the upper left and lower right corners of the rectangular area where the target is located and the probability value of the corresponding category. The system sets the confidence threshold to 0.5 to filter out low-confidence results, thereby outputting accurate target bounding box data.

[0044] Step S16: performing feature extraction processing on the target bounding box data and performing target classification to obtain target type information data, wherein the target type information data includes person target data, motion target data and text target data;

[0045] In this embodiment, the system extracts the image sub-region of each target area according to the target bounding box data obtained in step S15, and inputs it into a lightweight feature extraction network (such as ResNet or MobileNet) for further processing to extract the texture, shape and structural features in the area, and at the same time judge the motion state and static features of the target in combination with local background information; on this basis, the system uses a built-in classifier to make a comprehensive judgment on the extracted features and classify the target into a person, motion or text information, among which the person target may be confirmed by detecting facial features and posture information, the motion target depends on the displacement of the target in continuous frames, and the text target is identified by calling the optical character recognition module to identify the text that may exist in the area, and finally form the target type information data containing the detailed type identification of each target.

[0046] Step S17: performing coordinate calculation processing according to the target bounding box data and the target type information data, thereby obtaining target feature data including the target object type and position information.

[0047] In this embodiment, the system fuses the target bounding box data obtained in step S15 with the target type information data obtained in step S16, and calculates the center point coordinates of the rectangular area of ​​each target. The calculation method is to average the coordinates of the upper left corner and the lower right corner of the target rectangular area, respectively, to obtain the median in the horizontal and vertical directions, to ensure the accuracy of the center point position; then, the system integrates the center point coordinates, the bounding box size, and the target category identification into a unified data structure, and converts the relative coordinates into absolute coordinates according to the image resolution (for example, 1920×1080 pixels), and finally generates target feature data containing the target object type and precise position information, which provides key position information and target identification basis for subsequent image transmission processing steps such as thermal diffusion and weight calculation.

[0048] The present invention uses the image sensor of a digital camera to collect the light signal of the current scene, and generates initial pixel data through analog-to-digital conversion. This process is the starting point of image processing and ensures the complete collection of image information. Subsequently, the initial pixel data is subjected to white balance adjustment processing, which can correct the color deviation caused by different lighting conditions and make the image color more real and natural. The data after white balance correction is further subjected to color correction processing to further optimize the color performance of the image and ensure the accuracy and consistency of the color, which is crucial for subsequent target recognition and classification. On the basis of color correction, the image is subjected to denoising and sharpening processing, which can effectively remove the noise interference in the image and enhance the details and edge information of the image. The denoising processing reduces the random noise in the image, making the image clearer, while the sharpening processing highlights the outline and details of the image, making the features of the target object more obvious, and providing a higher quality image foundation for subsequent target detection. The target recognition processing based on the artificial intelligence target detection network can accurately identify the target bounding box in the image, which is a key step in image intelligent processing. Through the deep learning algorithm, the system can automatically identify various target objects in the image, such as people, moving objects and text, and generate a bounding box for each target. This process not only improves the efficiency of target recognition, but also greatly reduces the misrecognition rate, allowing the system to locate the target object more accurately. On the basis of target recognition, the target bounding box data is further subjected to feature extraction and classification processing, and the target object can be subdivided into different types such as human, sports and text. This classification processing enables the system to adopt different processing strategies according to different types of target objects, providing a basis for subsequent weight calculation and importance evaluation. For example, human targets may further analyze expression features, sports targets will analyze motion trajectories, and text targets will perform optical character recognition. Finally, through coordinate calculation processing, the target bounding box and target type information are combined to generate target feature data containing target object type and location information. This data not only contains the category information of the target, but also clarifies the specific location of the target in the image, providing precise guidance for subsequent image processing and analysis. This full-process optimization from image acquisition to target feature extraction ensures the efficiency and accuracy of image processing, enables the entire system to better adapt to complex image scenes, and provides a high-quality data foundation for subsequent image transmission and intelligent analysis.

[0049] Preferably, step S2 comprises the following steps:

[0050] Step S21: performing facial feature extraction processing on the person-type target in the target feature data, thereby obtaining expression feature data;

[0051] In this embodiment, the system extracts the facial area corresponding to the person target from the target feature data, and uses a pre-trained deep learning face detection algorithm (such as MTCNN or RetinaFace) to locate the face in the image. After positioning, key points are extracted for each detected facial area, and key points such as the eyes, nose tip, corners of the mouth and eyebrows are detected. The relative positions and shape characteristics of these parts are analyzed to form a feature vector describing the facial expression. The vector contains numerical information about subtle changes in facial expressions, such as the degree of upward movement of the corners of the mouth and the amplitude of opening and closing of the eyes. When processing an image with a resolution of 1920×1080 pixels, the entire process usually crops and scales the detected facial area to 64×64 pixels for subsequent rapid processing, thereby obtaining real-time and detailed expression feature data.

[0052] Step S22: performing emotion recognition processing on the expression feature data, when the expression feature shows a smile or a surprised expression, setting the importance weight coefficient of the character target object to 150% of the preset reference weight, otherwise the importance weight coefficient of the character target object is set to the preset reference weight, thereby obtaining the character emotion weight coefficient data;

[0053] In this embodiment, the system inputs the facial expression feature data obtained in S21 into an emotion recognition model constructed based on a convolutional neural network. After being trained with a large amount of labeled data, the model can accurately classify emotions such as smile, surprise, anger, and calmness. When the confidence of the smile or surprise emotion in the model output exceeds a set threshold (for example, 0.6), the system determines that the character target expresses positive emotions and sets its importance weight coefficient to 150% of the preset baseline weight, that is, multiplying the original weight by 1.5; if the condition is not met, the baseline weight remains unchanged, thereby generating emotion weight coefficient data corresponding to each character target. This method ensures that the emotional information can effectively weight the importance of the image in the subsequent processing stage.

[0054] Step S23: performing motion feature extraction processing on the motion target in the target feature data, thereby obtaining target motion trajectory data;

[0055] In this embodiment, the system uses a multi-frame target tracking algorithm to perform motion feature extraction processing on the part of the target feature data identified as a motion target. The specific operation is to extract the position information of the same target from consecutive image frames, calculate the displacement and speed change of the target between frames by applying the optical flow method (such as the Lucas-Kanade method) or a deep learning-based tracking network (such as SiamFC), and record the motion path of the target center point in consecutive frames, thereby forming motion trajectory data describing the motion state of the target. The data not only contains displacement information in the horizontal and vertical directions, but also reflects the continuity and change trend of the target motion. In a video stream with 30 frames per second, a 5-frame sliding window is usually used for smoothing to ensure the accuracy and real-time performance of the trajectory data.

[0056] Step S24: when the target motion trajectory data determines that the target motion state is an acceleration or turning action, the importance weight coefficient of the motion target object is set to 130% of the preset reference weight, otherwise the importance weight coefficient of the motion target object is set to the preset reference weight, thereby obtaining the motion target weight coefficient data;

[0057] In this embodiment, the system uses the target motion trajectory data calculated in S23 to judge the speed and direction changes of the target in continuous frames, and judges whether the target is in an acceleration or turning state by comparing the speed increment and direction change amplitude between adjacent frames (for example, the speed growth rate exceeds 20% or the direction change is greater than 15 degrees). If any of the above conditions is met, it is considered that the target motion state has changed significantly. At this time, the importance weight coefficient of the moving target object is set to 130% of the preset benchmark weight, that is, multiplying the basic weight by 1.3; if the conditions are not met, the basic weight is maintained, thereby obtaining the weight coefficient data for the moving target. This method can capture the key changes in the motion state in a timely manner and provide a basis for subsequent image processing.

[0058] Step S25: performing optical character recognition processing on the text-type targets in the target classification data, thereby obtaining text content data;

[0059] In this embodiment, the system uses optical character recognition (OCR) technology to process image areas identified as text-type targets in target classification data. First, the area is preprocessed, including binarization and noise removal (for example, the binarization threshold is determined using the Otsu method) to enhance the contrast between the text and the background. Then, a pre-trained OCR model (such as one based on CRNN or Tesseract deep learning version) is used to perform character segmentation and recognition on the preprocessed area to extract the characters and digital information in the area. This process can process text in different fonts and layouts. In standard scenarios (for example, the text area size is approximately 50×100 pixels), the recognition accuracy can reach more than 90%, thereby generating detailed text content data.

[0060] Step S26: performing importance matching processing on the text content data based on a preset rule library, thereby obtaining text target weight coefficient data;

[0061] In this embodiment, the system matches the text content data obtained in step S25 with a built-in preset rule library, which contains judgment rules for scores, time and other text information. The system determines the type of text content through string matching and regular expression detection. For example, if the text content conforms to the score format (such as containing numbers and dashes or colons), it is judged as score data, and the importance weight coefficient is set to 200% of the preset baseline weight. If the text content conforms to the time format (for example, it contains the ":" symbol and is preceded and followed by numbers), the weight is set to 180% of the baseline weight. In other cases, it is set to 120% of the baseline weight. The matching process can be completed within tens of milliseconds, ensuring that the weight coefficient data of the text target accurately reflects its importance in the specific application scenario.

[0062] Step S27: combining the text target weight coefficient data, the motion target weight coefficient data and the character emotion weight coefficient data into target importance data.

[0063] In this embodiment, the system integrates and fuses the character emotion weight coefficient data, motion target weight coefficient data and text target weight coefficient data obtained in the above steps according to the unique identification of the target in the image, and establishes a unified weight data record for each target object. If a target has multiple attributes at the same time, the system can adopt a weighted average or maximum value strategy to determine the final weight value, and finally generate target importance data including target category, location and comprehensive weight coefficient. This data structure provides a unified basic parameter for subsequent thermal diffusion processing, thereby realizing effective integration of multi-target information and real-time image processing in practical applications.

[0064] The present invention performs facial feature extraction processing on human targets, and can accurately capture the facial features of the person. This process can not only identify the basic facial features of the person, but also further analyze its emotional state. By performing emotion recognition processing on the facial features, the system can determine whether the person shows an expression with high emotional value such as a smile or surprise. When these specific emotions are detected, the importance weight coefficient of the human target is increased to 150% of the preset baseline weight, thereby ensuring that these emotionally rich areas are given priority in subsequent processing. This emotion-based weight adjustment enables image processing to better reflect the focus of human visual attention and enhances the emotional expression ability of the image. For motion targets, the system can obtain the motion trajectory data of the target through motion feature extraction processing. This data can reflect the motion state of the target, such as acceleration or turning. When the target shows these dynamic features, its importance weight coefficient is increased to 130% of the preset baseline weight, thereby ensuring that the key moments in the dynamic scene are better preserved. This sensitivity to the motion state enables the system to capture image areas with more visual impact and dynamic value, thereby improving the dynamic adaptability of image processing. For text targets, the system can accurately extract text content through optical character recognition processing. Text information often carries important semantic content, such as scores, time and other key data. By matching the importance of text content based on a preset rule library, the system can assign different weight coefficients according to the type and content of the text. For example, score data is assigned a higher weight to ensure that these important information is fully retained in image processing. This intelligent recognition and weight allocation of text content enables image processing to better adapt to complex scenes containing text information and enhances the semantic value of the image. Finally, the character emotion weight coefficient data, the motion target weight coefficient data and the text target weight coefficient data are combined into target importance data. This process realizes a comprehensive evaluation of the importance of different target objects in the image. Through this multi-dimensional weight calculation method, the system can comprehensively consider the emotional, dynamic and semantic information in the image, thereby providing more accurate and scientific guidance for subsequent image processing. This comprehensive weight evaluation system enables image processing to better adapt to the needs of diverse scenarios, improve the intelligence level of image processing, ensure that the image can retain the most critical information during transmission and storage, and meet users' needs for high-quality image processing.

[0065] Preferably, step S26 comprises the following steps:

[0066] The text content data is processed by importance matching based on the preset rule library to obtain text target weight coefficient data, wherein the preset rule library specifically sets the importance weight coefficient to 200% of the preset benchmark weight when the identified text target object is score data, and sets the importance weight coefficient to 180% of the preset benchmark weight when the identified text target object is time data; otherwise, the importance weight coefficient of the text target object is set to 120% of the preset benchmark weight.

[0067] In this embodiment, for the text content data extracted by the OCR module, the system first performs text preprocessing operations, including removing redundant spaces, unifying character formats (for example, converting full-width characters to half-width characters), and filtering noise information to ensure the standardization of the text to be matched; then, the system calls a built-in preset rule library, which contains multiple sets of matching modes and judgment criteria for different text data types, wherein the matching rule for score data is defined as that the text must contain two or more numbers at the same time, and the numbers are separated by dashes or colons (for example, "3-2" or "10:8"), while the matching rule for time data requires that a colon appears in the text, and the length of the numbers on both sides of the colon conforms to the general time format (for example, "08:15" or "12:30"); the system uses a string matching algorithm based on regular expressions to detect the preprocessed text content data one by one, first judging whether the text is a match or not. The system determines whether it conforms to the pattern of score data. If the match is successful, the text target is considered to be score data. At this time, the system sets its importance weight coefficient to twice the preset benchmark weight. For example, if the benchmark weight is 1.0, the weight of the score data is 2.0; if it does not meet the score rules, the system further detects whether it meets the rules of time data. If the match is successful, the weight coefficient is set to 1.8 times the preset benchmark weight; if both rules are not met, the system defaults to classifying the text target as other types and sets its weight coefficient to 1.2 times the preset benchmark weight; the entire matching process is usually completed within tens of milliseconds in common application scenarios (such as text area lengths between 50 and 200 characters), and the accuracy of recognition is guaranteed by setting matching confidence thresholds and multiple verification mechanisms. The final output text target weight coefficient data will serve as the key parameter for subsequent image importance weighting and heat map generation.

[0068] The preset rule base of the present invention classifies and assigns weights to the text content in detail. When the text target is identified as score data, the system will increase its importance weight coefficient to 200% of the preset benchmark weight. Score data has extremely high semantic value in many scenarios. For example, in sports events, the score is one of the most concerned information for the audience. By giving the score data a higher weight, the system can ensure that these key information are preferentially retained and highlighted during the image processing process, and the clarity and integrity of the score area can be guaranteed to the greatest extent even in image compression or other processing operations. For time data, the preset rule base sets its importance weight coefficient to 180% of the preset benchmark weight. Time information is also important in many scenarios. For example, in news reports, meeting records or event shooting, the timestamp can provide a clear time background for the image, helping users to quickly understand the specific time when the event occurred. By increasing the weight of time data, the system can ensure that this information will not be ignored during the image processing process, thereby better retaining the semantic integrity of the image. For other types of text content, although their importance is relatively low, they still have certain semantic value. Therefore, the system sets their importance weight coefficient to 120% of the preset benchmark weight. This weight setting not only ensures that the text information will not be over-compressed or lost, but also reasonably allocates processing priorities under limited resources to ensure the efficiency and quality of image processing. Through this importance matching processing based on a preset rule base, the system can flexibly adjust the weight coefficient according to the semantic value of different text content. This intelligent weight allocation mechanism makes image processing no longer a simple pixel operation, but can deeply understand the semantic information of the image and optimize the processing accordingly. During the image transmission and storage process, this method can ensure that key semantic information is better retained, while reasonably utilizing limited storage and transmission resources to improve the overall intelligence level of image processing. This precise evaluation and weight allocation of text content not only enhances the semantic expression ability of the image, but also provides users with image processing solutions that are more in line with actual needs.

[0069] Preferably, step S3 comprises the following steps:

[0070] Step S31: normalizing the target importance data, mapping the weight values ​​of all target objects to a range from 0 to 1, thereby obtaining standardized weight data;

[0071] In this embodiment, the system first uniformly collects the importance data of all targets, uses a traversal algorithm to scan the weight values ​​of all target objects, and determines the maximum and minimum values ​​in the current data set. Then, the principle of linear mapping is used to convert the weight value of each target from the original range (for example, the original weight may be between 1.0 and 2.0) to the standard range of 0 to 1. That is, for each target, the system subtracts the minimum value from its weight value and then divides it by the difference between the maximum and minimum values ​​to ensure that the lowest weight is mapped to 0 and the highest weight is mapped to 1. At the same time, when encountering the special case where all weight values ​​are equal, all values ​​are automatically set to 1 to avoid division by zero errors. This normalization process not only improves the robustness of subsequent processing, but also enables the importance of different targets to be compared on a unified scale. It is suitable for high-definition image data processing scenarios with 1920×1080 pixels. The entire process can be completed within tens of milliseconds with the support of the optimized algorithm.

[0072] Step S32: using Gaussian distribution method to perform thermal diffusion processing on the original image data based on the standardized weight data with the center point of the target object as the reference, so as to obtain single target thermal distribution data;

[0073] In this embodiment, the system generates a corresponding thermal diffusion distribution map for each target object using its normalized weight data. The specific operation is to first extract the coordinates of the target center point from the target feature data, and then calculate the diffusion radius based on the normalized weight value. A linear relationship is adopted, such as the diffusion radius increases by 10 pixels for every increase in the weight value of 0.1. The diffusion radius of the target is determined as the product of the normalized weight and a preset maximum diffusion radius (for example, 100 pixels). The diffusion radius is then used to generate a two-dimensional Gaussian kernel. The standard deviation of the Gaussian kernel is proportional to the diffusion radius. For example, the standard deviation increases by 1 unit for every increase in the diffusion radius of 20 pixels, so as to ensure that the generated Gaussian template has the highest thermal value at the center and gradually decays with the distance from the center. Finally, the Gaussian template is superimposed on the original image data with the target center as the center, and the single target thermal distribution data is generated through convolution operation. This method ensures that the influence range and thermal intensity of targets with different weights in the image have good visual consistency, which is suitable for the needs of real-time processing of video frames.

[0074] Step S33: performing overlapping area detection processing on the single target thermal distribution data of all targets, and marking the position information of the pixel points whose thermal values ​​are greater than a preset threshold, thereby obtaining thermal overlapping area data;

[0075] In this embodiment, the system performs pixel-by-pixel detection on all single-target thermal distribution data, and adopts a regional growing algorithm and a threshold determination method. First, the thermal distribution data of each target are superimposed in space, and the cumulative thermal value of each pixel is detected to see whether it exceeds a preset threshold (for example, the threshold is set to 0.5). If it exceeds, the position information of the pixel is recorded and marked as a thermal overlap area. At the same time, by scanning adjacent pixels and using a connectivity analysis algorithm to determine the boundaries of local high-value thermal areas, one or more thermal overlap area data structures are formed. This process can accurately capture the intersecting parts of multiple target thermal distributions and independently identify these areas, providing accurate regional information for subsequent maximum value selection processing. The entire detection process can be quickly completed in real-time applications in typical application scenarios such as 1920×1080 pixel images after efficient algorithm optimization.

[0076] Step S34: performing maximum value selection processing on the thermal overlap area data to obtain thermal superposition data, wherein the maximum value selection processing selects the maximum value of all overlapping thermal values ​​as the final thermal value of the pixel point in the overlap area;

[0077] In this embodiment, the system processes the pixel points in each overlapping area according to the thermal overlapping area data marked in the previous step, and adopts a point-by-point comparison algorithm to collect the thermal values ​​of all target thermal distribution data at each pixel position in the overlapping area. Then, these thermal values ​​are compared and the maximum value is selected as the final thermal value of the pixel, thereby avoiding the thermal oversaturation problem that may be caused by simple superposition. This maximum value selection process not only ensures that the most significant thermal effect in the overlapping area is retained, but also balances the data smoothness and local details of the overall thermal map, which is suitable for fine layered display of multiple targets in complex scenarios.

[0078] Step S35: Perform global normalization processing on the thermal superposition data to obtain importance heat map data.

[0079] In this embodiment, the system uses the thermal superposition data obtained by maximum value selection processing as the basis of the overall thermal map, and then uses a global normalization algorithm to process the entire thermal map data. The specific method is to first scan the thermal data of the entire image to find the global highest thermal value and the lowest thermal value, and then use a linear normalization method to scale the thermal value of each pixel point to the range of 0 to 1, so that all thermal values ​​fall within the standard range, thereby making the thermal distribution in the image more balanced and convenient for subsequent display or analysis. The global normalization process ensures that no matter how the absolute size of the thermal value in the original data changes, stable and intuitive thermal map data can be output, which is suitable for various real-time image transmission and data visualization scenarios.

[0080] The present invention normalizes the target importance data and maps the weight values ​​of all target objects to a range of 0 to 1. This process ensures that the importance weights of different target objects can be compared and processed on the same scale. The normalization process eliminates the dimensional differences between the weight values, so that the system can more fairly evaluate the importance of each target object, and provides a standardized data basis for subsequent thermal diffusion processing. Subsequently, the Gaussian distribution method is used for thermal diffusion processing, and the importance weight of each target is diffused to the pixel area around it with the center point of the target object as the reference. This diffusion method based on Gaussian distribution conforms to the characteristics of human visual perception and can naturally simulate the range of human eye attention to important targets. The single-target thermal distribution data generated in this way can not only highlight the importance of the target object itself, but also take into account the correlation of its surrounding areas, so that the thermal map can more realistically reflect the visual influence of the target object in the image. When processing multiple target objects, the system will perform overlapping area detection on the single-target thermal distribution data of all targets, and mark the pixel positions whose thermal values ​​are greater than the preset threshold. This process can identify the areas of mutual influence between multiple target objects and ensure that the thermal values ​​in the overlapping areas can be reasonably processed. Through the maximum value selection process, the system will select the maximum value of all overlapping thermal values ​​as the final thermal value of the pixel. This processing method can ensure that when multiple target objects overlap, the heat map can accurately reflect the highest importance of each pixel, avoiding the underestimation or loss of thermal values. Finally, the thermal superposition data is globally normalized to obtain the final importance heat map data. This global normalization step can ensure the overall consistency of the heat map, making the distribution of thermal values ​​in the heat map more uniform and reasonable. The importance heat map generated in this way can not only intuitively reflect the importance of each area in the image, but also provide an accurate reference for subsequent image layering compression and other processing. This heat map generation process from local to global enables the system to better adapt to complex image scenes, ensure the intelligence and efficiency of image processing, and provide users with a more scientific and reasonable image processing solution.

[0081] Preferably, step S32 includes the following steps:

[0082] Step S321: Calculate the thermal diffusion radius of the target object in the original image data according to the standardized weight data, so as to obtain diffusion radius data, wherein the diffusion radius is proportional to the standardized weight value, and the diffusion radius increases by 10 pixels for every increase of 0.1 in the standardized weight value;

[0083] In this embodiment, the system first reads the standardized weight data of each target object, and the data value is in the range of 0 to 1, and then calculates the thermal diffusion radius using a preset proportional relationship, that is, the system multiplies the standardized weight value of each target by a fixed incremental ratio. The specific method is that when the standardized weight value increases by 0.1, the diffusion radius increases by 10 pixels. Therefore, for a target with a standardized weight value of 0.3, its diffusion radius will be calculated as 30 pixels, and for a target with a weight value of 0.8, its diffusion radius is 80 pixels. This linear proportional method ensures that the importance of the target directly affects the diffusion range, which is suitable for the needs of personalized thermal diffusion for different targets in high-definition images (such as 1920×1080 pixels). At the same time, the system will smooth the edge values ​​(such as weights close to 0 or 1) during the processing process to avoid the problem of too small or too large diffusion radius in extreme cases.

[0084] Step S322: performing two-dimensional Gaussian kernel generation processing on the diffusion radius data to obtain Gaussian diffusion template data, wherein the standard deviation of the Gaussian kernel is proportional to the diffusion radius, and the standard deviation increases by 1 unit for every 20 pixels increase in the diffusion radius;

[0085] In this embodiment, the system generates a two-dimensional Gaussian kernel using the diffusion radius data obtained in S321. The specific operation is to determine the standard deviation of the Gaussian kernel according to the diffusion radius of each target, and adopt a preset proportional relationship, that is, for every 20 pixels increase in the diffusion radius, the standard deviation increases by 1 unit. Therefore, if the diffusion radius of a target is 40 pixels, the corresponding standard deviation is 2 units, and when the diffusion radius is 100 pixels, the standard deviation is 5 units; on this basis, the system generates a two-dimensional Gaussian distribution template, the size of which is usually selected to be 6 times the standard deviation plus 1 (to cover a sufficient probability density range), and by calculating the distance of each pixel point in the template relative to the center point, the corresponding weight value is assigned according to the Gaussian function attenuation law. This process ensures that the generated Gaussian diffusion template can accurately reflect the thermal distribution characteristics of the target attenuating from the center to the outside in the image, and provide a fine diffusion model for subsequent convolution operations. At the same time, in practical applications, the template size can be dynamically adjusted according to image resolution and processing speed requirements.

[0086] Step S323: performing target center point positioning processing on the original image data according to the position information in the target feature data, thereby obtaining heat diffusion center point data;

[0087] In this embodiment, the system performs center point positioning processing on the target in the original image data based on the position information recorded in the target feature data. The specific operation is to extract the coordinate information of the upper left corner and the lower right corner from the bounding box data of the target, and calculate the average value of the two coordinates to determine the geometric center point of the target, that is, the horizontal coordinate is half of the sum of the left and right coordinates, and the vertical coordinate is half of the sum of the upper and lower coordinates. The obtained center point data is accurately mapped to the pixel grid of the original image (for example, in a 1920×1080 pixel image, the x value of the center point coordinate is between 0 and 1919, and the y value is between 0 and 1079). This positioning method ensures that the thermal diffusion process is always centered on the core area of ​​the target. If the target is located at the edge of the image, the system will automatically handle the boundary truncation to ensure the accuracy of the center point data in subsequent convolution operations.

[0088] Step S324: using the Gaussian diffusion template data to perform convolution operation on the thermal diffusion center point data, so as to obtain single target thermal distribution data.

[0089] In this embodiment, the system performs convolution operation on the Gaussian diffusion template generated in S322 and the target center point data obtained in S323. The specific operation is to position the Gaussian diffusion template according to the position of the target center point in the original image, map the weight value of each pixel in the template to the corresponding pixel position of the original image, and superimpose the weight distribution of the template on a blank thermal map through a convolution operation, thereby forming single-target thermal distribution data. This process performs weighted summation on all pixels in the target area so that the thermal value at the target center point is the highest, and the thermal value gradually decays in a Gaussian distribution with increasing distance from the center. In actual applications, if the Gaussian template partially exceeds the image boundary, the system will use boundary processing techniques such as mirroring or zero filling to ensure the continuity and accuracy of the operation, and use hardware acceleration and optimization algorithms to realize real-time convolution operations. The final output single-target thermal distribution data can be used to generate a comprehensive thermal map and guide subsequent image layered compression processing.

[0090] The present invention calculates and processes the thermal diffusion radius, and the system determines a suitable diffusion radius for each target object according to the standardized weight data. This diffusion radius is proportional to the standardized weight value. The higher the weight, the larger the diffusion radius, which means that the importance and influence range of the target object are wider. This method of dynamically adjusting the diffusion radius can ensure that the thermal diffusion range of important target objects matches their actual importance, and avoids the unreasonable thermal distribution that may be caused by a fixed diffusion radius. For example, for targets with higher weights, such as key figures or important text information, a larger diffusion radius can better highlight their visual importance in the image. Subsequently, the diffusion radius data is processed by two-dimensional Gaussian kernel generation to generate Gaussian diffusion template data. The standard deviation of the Gaussian kernel is proportional to the diffusion radius. This design enables thermal diffusion to more naturally simulate the characteristics of human visual perception. The smoothing characteristics of the Gaussian kernel can ensure that the thermal value gradually decays during the diffusion process, rather than suddenly changing, thereby generating a thermal distribution that is more in line with visual habits. This diffusion method based on the Gaussian kernel can not only highlight the central area of ​​the target object, but also reasonably extend its influence range, so that the heat map can more realistically reflect the importance distribution of the target object in the image. After determining the diffusion radius and Gaussian kernel, the system performs target center point positioning processing on the original image data according to the position information in the target feature data to obtain the thermal diffusion center point data. This process ensures that the starting point of thermal diffusion is accurate, so that the thermal distribution can be accurately carried out around the target object. Accurate target center point positioning is the key to generating high-quality thermal maps. It can ensure that thermal diffusion does not deviate from the actual position of the target object, thereby improving the accuracy and reliability of the thermal map. The thermal diffusion center point data is processed by convolution operation using Gaussian diffusion template data to obtain single-target thermal distribution data. Convolution operation is an efficient image processing technology that can combine Gaussian diffusion template with target center point to generate a smooth and continuous thermal distribution. This thermal distribution can not only highlight the importance of the target object itself, but also reasonably diffuse to its surrounding area to form a natural thermal field. The single-target thermal distribution data generated in this way provides high-quality basic data for subsequent thermal map synthesis and image processing. By dynamically adjusting the diffusion radius, generating Gaussian kernels, accurately locating the target center point, and performing convolution operations, the generation process of the thermal map is ensured to be both scientific and efficient. This heat diffusion method based on target importance can provide more accurate guidance for image processing, so that the image can better retain key information in subsequent compression or other processing processes, while optimizing resource allocation and improving the quality and efficiency of overall image processing.

[0091] Preferably, step S33 includes the following steps:

[0092] Step S331: establishing a pixel-level thermal value matrix according to the single-target thermal distribution data, and performing non-zero judgment processing on each pixel point in the matrix, thereby obtaining effective thermal area data;

[0093] In this embodiment, the system first uses the thermal distribution data of a single target to construct a pixel-level thermal value matrix. The specific method is to arrange the thermal distribution data of each target into a two-dimensional matrix according to its corresponding position in the original image (for example, a matrix containing 2073600 elements is constructed in an image with a resolution of 1920×1080). The system then traverses the matrix and makes a non-zero judgment on the thermal value of each pixel, that is, to determine whether the thermal value of the pixel is greater than zero. All pixels greater than zero are marked as valid thermal points, while zero-value pixels are excluded, thereby forming a data set containing only valid thermal areas. This process can accurately extract key information of the target thermal distribution area while maintaining a high processing speed, and record the specific position and thermal value of each valid point, laying the foundation for subsequent regional analysis.

[0094] Step S332: extracting the region boundary of the effective thermal region data, determining the boundary contour of each target thermal distribution, and thus obtaining thermal boundary data;

[0095] In this embodiment, based on the effective thermal area data obtained in step S331, the system uses a method combining an edge detection algorithm with a connected region analysis to extract the boundary contour of each target thermal distribution. The specific operation is to first use a gradient detection method (such as a Sobel operator) to scan the grayscale changes of pixels in the effective thermal area, mark the pixels with large gradient changes as edge candidate points, and then combine these candidate points into a continuous boundary contour through a connected component analysis algorithm, and finally obtain the external boundary data of each target thermal area. These boundary data record the contour information of each target in the form of a coordinate set, which is suitable for accurately segmenting the target area under complex backgrounds and providing necessary geometric data support for spatial relationship analysis.

[0096] Step S333: Calculate the spatial relationship between the thermal regions of different targets according to the thermal boundary data, and determine whether there is an intersection between the boundaries according to the diffusion radius data, thereby obtaining the region intersection data;

[0097] In this embodiment, the system uses the thermal boundary data obtained in step S332 to calculate the spatial relationship of the thermal areas of different targets. The specific method is that the system compares the boundary contours of each target area, and determines whether there is an intersection between the thermal areas of each target by calculating the shortest distance between the boundaries and the degree of overlap. At the same time, combined with the diffusion radius data of each target, if the distance between two targets is less than the sum of their diffusion radii, it is considered that the two areas have an intersection. The system records all detected area intersections and generates a set of area intersection data containing the intersection area location, participating target identification and overlap degree information, to ensure that subsequent processing can accurately locate and analyze multi-target overlap.

[0098] Step S334: performing positioning and marking processing on the pixel points in the region intersection data, and assigning a unique position index to the pixel points in each overlapping region, thereby obtaining overlapping pixel index data;

[0099] In this embodiment, the system performs pixel-level positioning and marking processing on each intersection area in the area intersection data obtained in step S333. The specific operation is that the system first traverses all pixel points in the intersection area, and assigns a unique position index to each pixel according to its row and column coordinates in the original image. Usually, an incremental counter method is used to generate a unique number for each pixel to ensure that the pixel points in each overlapping area can be accurately distinguished and tracked, and finally a detailed overlapping pixel index data is formed. This data set not only contains pixel position and index information, but also provides a direct reference basis for subsequent thermal value mapping and regional statistical analysis.

[0100] Step S335: extracting the thermal value in each overlapping area according to the overlapping pixel index data, and establishing a mapping relationship between the thermal value and the pixel position, thereby obtaining thermal mapping data;

[0101] In this embodiment, the system uses the overlapping pixel index data obtained in step S334 to extract the specific thermal value of each pixel point corresponding to the index in the original thermal distribution data, and then maps the thermal value to the specific position of the pixel point in the image one by one to form a structured thermal mapping data table. The table uses a key-value pair storage method, where the key is the pixel position index and the value is the corresponding thermal intensity and its row and column coordinate information. This mapping relationship not only facilitates the subsequent detailed analysis of the thermal distribution in the overlapping area, but also provides accurate data support for the visualization and data statistics of the thermal map, ensuring that the multi-target thermal information is reasonably displayed in the entire image.

[0102] Step S336: performing threshold comparison processing on the thermal value of each pixel in the thermal mapping data, screening out the pixel whose thermal value is greater than the preset threshold, thereby obtaining high thermal pixel data;

[0103] In this embodiment, the system performs pixel-by-pixel threshold comparison processing on the thermal mapping data constructed in step S335. The specific method is that the system sets a preset threshold (for example, set to 0.7 within the normalized range), traverses the thermal value of each pixel in the thermal mapping data, and filters out those pixels whose thermal values ​​are higher than the preset threshold, and records their related thermal values ​​and corresponding pixel positions as high thermal pixel data. This process quickly locates all high thermal areas in the entire image through an efficient threshold comparison algorithm, ensuring that these areas with higher visual weights can be focused on during subsequent image layering or compression processing.

[0104] Step S337: Associating the overlapping pixel index data and the high thermal pixel data to obtain thermal overlapping area data.

[0105] In this embodiment, the system associates the overlapping pixel index data generated in step S334 with the high thermal pixel data screened out in step S336. The specific operation is that the system matches the two data sets through a unique position index, finds out the pixels that are marked as overlapping pixels and have high thermal values ​​in the same overlapping area, and groups and summarizes these pixels by area, and finally generates a complete thermal overlapping area data. This data not only records in detail the distribution and number of high thermal pixels in each overlapping area, but also provides statistical information on the overall thermal intensity of the area, providing an accurate and reliable basis for subsequent image compression and target importance analysis.

[0106] The present invention establishes a pixel-level thermal value matrix and performs non-zero judgment processing on each pixel in the matrix, so that the system can quickly screen out areas with effective thermal values. This process not only removes blank areas in the thermal distribution, but also provides a clear data basis for subsequent boundary extraction and regional analysis. The generation of effective thermal area data enables the system to focus on those pixels that really have important information, thereby improving processing efficiency and accuracy. The effective thermal area data is subjected to regional boundary extraction to determine the boundary contours of each target thermal distribution. This process can clearly outline the scope of each target thermal area, so that the system can clearly distinguish the thermal influence range of different target objects. The extraction of thermal boundaries provides an important reference basis for subsequent spatial relationship analysis, ensuring that the mutual relationship between thermal areas can be accurately identified. Based on the thermal boundary data, the system further calculates the spatial relationship between different target thermal areas, and determines whether there is an intersection between the boundaries based on the diffusion radius data. This process can identify the overlap between thermal areas, providing key information for subsequent thermal value processing. By judging the intersection of boundaries, the system can accurately identify which areas are affected by multiple target objects, thus providing a scientific basis for the merging of thermal values. After identifying the overlapping areas, the system performs positioning and marking processing on the pixels in the intersection of the areas and assigns a unique position index to each pixel. This process ensures that each pixel in the overlapping area can be accurately identified and tracked, providing accurate positioning information for subsequent thermal value extraction and processing. The generation of overlapping pixel index data enables the system to perform fine processing on the overlapping areas, avoiding confusion and loss of thermal values. The system extracts the thermal values ​​in each overlapping area based on the overlapping pixel index data, and establishes a mapping relationship between the thermal values ​​and the pixel positions. The thermal mapping data generated by this process enables the system to clearly understand the thermal value of each pixel and its corresponding position, providing important data support for subsequent thermal value comparison and screening. By performing threshold comparison processing on the thermal values ​​in the thermal mapping data, the system can screen out pixels whose thermal values ​​are greater than the preset threshold, thereby obtaining high thermal pixel data. This process further highlights the important areas in the thermal distribution, ensuring that subsequent processing can focus on these pixels with high thermal values, thereby better retaining the key information in the image. The overlapping pixel index data is associated with the high thermal pixel data to generate thermal overlap area data. This process not only integrates the thermal information in the overlapping area, but also provides complete data support for subsequent thermal map synthesis. In this way, the system can ensure that when the thermal distribution of multiple target objects overlaps, the thermal values ​​can be accurately processed and retained, avoiding the loss of information or incorrect superposition. The accuracy and reliability of the thermal map are ensured through precise boundary extraction, spatial relationship analysis, pixel positioning, thermal value extraction and threshold screening.This refined processing of thermal overlapping areas enables the system to better adapt to complex image scenes and provides a high-quality data foundation for subsequent image processing, thereby improving the intelligence level and processing effect of the entire image processing system.

[0107] Preferably, step S4 comprises the following steps:

[0108] Step S41: dividing the original image data into image blocks of equal size according to the importance heat map data, and calculating the average heat value of each image block, thereby obtaining the image block heat data;

[0109] In this embodiment, the system first divides the original image data into image blocks of equal size according to a fixed size based on the pre-generated importance heat map data. For example, in a 1920×1080 pixel image, it can be divided into grid areas of 16×16 pixels each to ensure that each image block has a consistent number of pixels; then, the system traverses each image block, accumulates the thermal values ​​of all pixels in the block, and divides the accumulated value by the total number of pixels in the image block to calculate the average thermal value of the image block. This average value represents the overall thermal level of the area and is recorded as the image block thermal data. This process uses a simple weighted average method to ensure the efficiency and accuracy of data calculation. At the same time, in actual applications, the size parameters of the image block can be adjusted according to the image resolution and processing requirements to balance the requirements of detail retention and processing speed.

[0110] Step S42: performing hierarchical processing on the image block thermal data based on a preset first threshold and a preset second threshold, thereby obtaining hierarchical image block data, wherein the hierarchical image block data includes a high importance block, a medium importance block, and a low importance block;

[0111] In this embodiment, the system uses two preset thresholds to perform hierarchical processing on the thermal data of the image blocks obtained in step S41, wherein the preset first threshold and the second threshold can be set to 0.7 and 0.4 respectively (assuming that the thermal value has been normalized to between 0 and 1), and the system compares the average thermal value of each image block: if the value is higher than the first threshold, the image block is classified as a high importance block; if the value is between the first threshold and the second threshold, it is classified as a medium importance block; and the image block below the second threshold is judged to be a low importance block, forming hierarchical image block data. The hierarchical strategy enables different compression ratios and encoding strategies to be adopted for different areas in the subsequent compression process, thereby achieving an effective reduction in data volume while ensuring key details. The threshold setting can be adjusted according to the scene and user needs in actual applications to optimize the balance between visual quality and transmission efficiency.

[0112] Step S43: performing hierarchical compression processing according to different importance levels of the hierarchical image block data, thereby obtaining high-importance compressed data, medium-importance compressed data, and low-importance compressed data;

[0113] In this embodiment, the system adopts customized compression processing strategies according to the image block data of different importance levels obtained in step S42. For high-importance blocks, the system adopts lossless compression encoding methods (such as PNG or lossless JPEG) to ensure that the image details are fully preserved; for medium-importance blocks, the system adopts JPEG compression processing with a medium compression ratio, and its compression quality parameters are linearly mapped and adjusted according to the average thermal value of the image block. For example, a higher quality parameter (such as about 80) is used in areas with higher thermal values, and a slightly lower quality parameter is used in areas with lower thermal values, so as to take into account both details and compression rate; for low-importance blocks, a high compression ratio is used, and the compression ratio may be further improved by increasing the quantization step size (for example, the compression quality parameter is set to about 50). This method implements a dynamic encoding strategy based on the importance of different areas, which not only ensures the visual effect of key areas, but also effectively reduces the overall data volume, and meets different network bandwidth and storage requirements through parameter adjustment in actual scenarios.

[0114] Step S44: organize and process the high-importance compressed data, the medium-importance compressed data, and the low-importance compressed data, and record the compression type and parameter information of each image block in the data header, so as to obtain layered compressed image data.

[0115] In this embodiment, the system uniformly organizes and processes the high-importance compressed data, medium-importance compressed data, and low-importance compressed data generated in the aforementioned step S43. The specific method is to arrange the compressed data of each image block according to its spatial position in the original image, and add detailed metadata to the data header. The metadata includes a compression type identifier for each image block (for example, marked as lossless, medium, or high compression), the compression parameters used (such as JPEG quality parameters, quantization step values, etc.), and a position index of the image block in the entire image, to ensure that the image data of each image block can be accurately restored based on this information during subsequent decoding. The entire organization process adopts a structured data encapsulation format (such as a custom data packet or a standard container format) to support fast parsing and transmission, which is suitable for real-time application scenarios that require layered compressed image transmission, and achieves efficient storage and transmission while ensuring data integrity.

[0116] The present invention divides the original image into image blocks of equal size according to the importance heat map data, and calculates the average heat value of each image block. This process provides a quantitative basis for subsequent hierarchical processing, so that the system can classify image blocks according to the high and low heat values. In this way, the image is decomposed into multiple regions with clear importance identification, providing accurate guidance for subsequent compression strategies. Based on the preset first threshold and second threshold, the system performs hierarchical processing on the image block heat data, and divides the image blocks into high importance blocks, medium importance blocks and low importance blocks. This hierarchical method can ensure that the system can distinguish the importance of different regions when processing images, so as to adopt different strategies in the subsequent compression process. High importance blocks usually contain key information, such as human expressions, important text or dynamic targets, and their details need to be retained first; while low importance blocks can accept higher compression ratios, thereby saving storage space and transmission bandwidth. On the basis of hierarchical processing, the system performs hierarchical compression processing according to the different importance levels of image blocks. High-importance blocks use lossless compression to ensure the integrity and high quality of key information; medium-importance blocks use medium compression ratios to balance image quality and storage requirements; low-importance blocks use high compression ratios to further optimize storage and transmission efficiency. This differentiated compression strategy enables the system to retain key information of the image as much as possible under limited resources, while reducing redundant data and improving overall processing efficiency. The system organizes and processes the hierarchically compressed data and records the compression type and parameter information of each image block in the data header. This process not only makes the compressed image data well structured, but also provides necessary information for subsequent decompression and image reconstruction. By recording the compression type and parameters, the system can accurately restore the original state of the image block during decompression to ensure the integrity and consistency of the image. This series of hierarchical compression processing steps achieves efficient and intelligent image processing through precise image block division, reasonable hierarchical strategies, differentiated compression methods, and structured data organization. This hierarchical compression method can not only significantly reduce the storage and transmission requirements of image data, but also ensure that key information of the image is retained first, thereby improving the overall performance and user experience of the system while meeting the user's image quality requirements.

[0117] Preferably, step S43 includes the following steps:

[0118] Step S431: performing lossless compression encoding processing on high-importance blocks in the layered image block data, thereby obtaining high-importance compressed data;

[0119] In this embodiment, the system extracts all high-importance blocks from the layered image block data. These blocks usually represent the most critical detail areas in the image, and then uses a lossless compression coding method to process each high-importance block. The specific operations include predictive coding of the pixel data in the block (for example, using intra-row or inter-row prediction) to eliminate redundant information, and entropy coding the prediction residual using a coding algorithm such as PNG or lossless JPEG to ensure that the compressed data can completely restore the original pixel information when decoded; in actual application scenarios, for each 16×16 pixel high-importance block, the compression level parameter can be set to 6 (in the range of 0 to 9) to obtain a more ideal compression ratio while ensuring that the key details of the image are not lost. The high-importance compressed data finally generated will be stored in the layered compressed image data in the form of a data stream.

[0120] Step S432: setting JPEG compression quality parameters for the medium importance block in the layered image block data, thereby obtaining medium compression parameter data, wherein the JPEG compression quality parameter setting is specifically to linearly map the compression quality parameters according to the thermal value of the medium importance block;

[0121] In this embodiment, when the system processes the medium-importance blocks in the layered image block data, it first calculates the average thermal value in each block, and then converts the thermal value into a JPEG compression quality parameter based on a preset linear mapping relationship. The mapping relationship is usually set so that when the thermal value of the block is in a preset medium range (for example, between 0.4 and 0.7), the higher the thermal value, the higher the corresponding JPEG quality parameter. For example, if the quality parameter is set to 80 when the thermal value is 0.7 and to 60 when the thermal value is 0.4, a block with a thermal value of 0.55 will obtain a quality parameter of approximately 70. The mapping process uses a simple linear formula to convert the thermal value into a compression quality parameter suitable for the JPEG algorithm, and records the obtained medium compression parameter data so that customized JPEG compression processing can be performed on the medium-importance blocks later to ensure an ideal balance between the details and compression rate of relatively important areas in the image.

[0122] Step S433: performing JPEG compression processing on the medium importance block in the layered image block data according to the medium compression parameter data, thereby obtaining medium importance compressed data;

[0123] In this embodiment, the system performs JPEG compression processing on each medium importance block according to the medium compression parameter data obtained in step S432. The specific operations include first converting each image block into a YCbCr color space, then performing an 8×8 discrete cosine transform (DCT) on the data in the block, and then quantizing the DCT coefficients using a quantization table determined according to the medium compression parameter data. The setting of each value in the quantization table is directly affected by the quality parameter. When the quality parameter is high, a lower quantization factor is used to retain more details. When the quality parameter is low, a higher quantization factor is used to improve the compression ratio. Subsequently, entropy coding (such as Huffman coding or arithmetic coding) is performed on the quantized coefficients to finally generate medium importance compressed data. This process can be completed within tens of milliseconds in typical real-time image transmission applications, thereby significantly reducing the data volume while ensuring the visual quality of the medium area.

[0124] Step S434: performing a quantization step index mapping process based on the average thermal value of each block on the low-importance blocks in the layered image block data, thereby obtaining low-compression parameter data, wherein the quantization step index mapping process specifically includes exponentially increasing the quantization step as the thermal value decreases;

[0125] In this embodiment, the system first calculates the average thermal value of each block of low-importance blocks in the layered image block data, and then adopts a quantization step determination method based on exponential mapping. According to preset rules, when the average thermal value of the block is low, the system increases the quantization step in an exponential manner to further improve the compression ratio. For example, the basic quantization step is set as a benchmark. When the thermal value is lower than 0.4, the quantization step is set to the benchmark step multiplied by an exponential factor through a formula (such as the quantization step increases by a fixed exponential multiple for every 0.1 decrease in the thermal value), thereby achieving a higher compression rate in visually less important areas. Finally, the calculated quantization step values ​​of each low-importance block are organized into low-compression parameter data. This mapping process ensures that the amount of data in low-importance areas can be greatly reduced through a larger quantization step while having little impact on the overall visual effect.

[0126] Step S435: Perform high compression ratio compression processing on low-importance blocks in the layered image block data according to the low-compression parameter data, thereby obtaining low-importance compressed data.

[0127] In this embodiment, the system performs high compression ratio compression processing on the low-importance blocks according to the low-compression parameter data generated in step S434. The specific operation process is similar to JPEG compression, but a larger quantization step size determined in step S434 is used in the quantization stage, so that the DCT coefficients of the low-importance blocks produce more zero coefficients after a rougher quantization process, and then a higher compression ratio is achieved in the entropy coding stage. The process includes first performing a discrete cosine transform on the image block, then quantizing the DCT coefficients using a pre-calculated exponential mapping quantization step size, and finally encoding the quantization result into a data stream using an efficient entropy coding algorithm, and finally generating low-importance compressed data. This method is particularly suitable for video surveillance or network transmission scenarios, and can achieve significant data compression for background areas with less visual impact to meet the requirements for fast transmission in bandwidth-constrained environments.

[0128] For high-importance image blocks, the system of the present invention adopts lossless compression coding processing. This processing method can ensure that the most critical information in the image, such as human expressions, important text or moving targets, will not lose any details during the compression process. Although lossless compression does not reduce the redundancy of image data, it can store these key information in an efficient way, so that the quality of the original image can be completely restored in the subsequent decompression process. This fine processing of high-importance blocks ensures that the core value of the image is completely preserved. For medium-importance image blocks, the system adopts JPEG compression and linearly maps the compression quality parameters according to the thermal value. This strategy can dynamically adjust the compression quality according to the actual importance of the image block. Medium-importance blocks with higher thermal values ​​will adopt higher compression quality parameters to retain more details; while medium-importance blocks with lower thermal values ​​will appropriately reduce the compression quality parameters, thereby further optimizing the storage space under the premise of ensuring the basic quality of the image. This dynamic adjustment based on thermal values ​​enables the system to find the best balance between different importance levels, saving storage resources and avoiding the degradation of image quality caused by excessive compression. For low-importance image blocks, the system uses high compression ratio compression processing and determines the compression parameters through quantization step index mapping processing. This processing method is particularly targeted at those areas that have little impact on the overall image quality. The quantization step index mapping processing makes the quantization step of the image block with lower thermal value increase exponentially, thereby achieving a higher compression ratio. This strategy can minimize the amount of data of low-importance image blocks without significantly affecting the visual effect, thereby freeing up more storage and transmission resources for high-importance and medium-importance image blocks. Through this hierarchical and differentiated compression processing, the system can flexibly adjust the compression strategy according to the importance of different image blocks. The lossless compression of high-importance blocks ensures the complete retention of key information, the dynamic JPEG compression of medium-importance blocks balances quality and storage requirements, and the high compression ratio processing of low-importance blocks further optimizes the overall resource utilization efficiency. This refined compression method not only improves the intelligence level of image processing, but also ensures that images can be presented to users with the best quality under limited storage and transmission resources, thereby significantly improving the performance and user experience of the entire image processing system.

[0129] The present invention also includes an image transmission system for a digital camera, which is used to execute the above-mentioned image transmission method for a digital camera. The image transmission system for a digital camera includes:

[0130] The image acquisition module is used to obtain the original image data taken by the digital camera; perform target scanning processing on the original image data to obtain target feature data including target object type and position information;

[0131] A target weight calculation module is used to perform weight calculation processing on each target object according to the target feature data to obtain target importance data, wherein the weight calculation processing specifically includes determining the weight value based on the facial expression features of the person when the target object type is a person, determining the weight value based on the motion trajectory when the target object type is a moving target, and determining the weight value based on a preset rule when the target object type is text information;

[0132] A heat map generation module is used to perform heat diffusion processing on the original image data based on the target importance data using a Gaussian distribution method with the center point of the target object as a reference to obtain importance heat map data, wherein the heat diffusion processing uses a maximum value superposition method when the heat distributions of multiple target objects overlap;

[0133] The image layered compression module is used to perform layered compression processing on the original image data according to the importance heat map data to obtain layered compressed image data, wherein the layered compression processing is specifically to perform lossless compression on the area where the thermal value is greater than a preset first threshold, to perform medium compression ratio compression on the area where the thermal value is between the first threshold and the preset second threshold, and to perform high compression ratio compression on the area where the thermal value is less than the second threshold.

[0134] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0135] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An image transmission method for a digital camera, characterized in that: The following steps are involved: Step S1: acquiring original image data taken by a digital camera; performing target scanning processing on the original image data to obtain target feature data including target object type and position information; Step S2: performing weight calculation processing on each target object according to the target feature data to obtain target importance data, wherein the weight calculation processing specifically includes determining the weight value based on the expression feature of the person when the target object type is a person, determining the weight value based on the motion trajectory when the target object type is a moving target, and determining the weight value based on a preset rule when the target object type is text information; Step S3: using Gaussian distribution method to perform thermal diffusion processing on the original image data based on the target importance data with the center point of the target object as the reference to obtain importance heat map data, wherein the thermal diffusion processing uses a maximum value superposition method when the thermal distributions of multiple target objects overlap; Step S4: Performing layered compression processing on the original image data according to the importance heat map data to obtain layered compressed image data, wherein the layered compression processing is specifically performing lossless compression on areas where the thermal value is greater than a preset first threshold, performing medium compression ratio compression on areas where the thermal value is between the first threshold and a preset second threshold, and performing high compression ratio compression on areas where the thermal value is less than the second threshold.

2. The image transmission method for a digital camera according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: using the image sensor of the digital camera to collect the light signal of the current scene, and generating initial pixel data through analog-to-digital conversion; Step S12: performing white balance adjustment processing on the initial pixel data to obtain white balance correction data; Step S13: performing color correction processing on the white balance correction data to obtain color correction image data; Step S14: performing image noise reduction processing and image sharpening processing according to the color-corrected image data, thereby obtaining original image data; Step S15: performing target recognition processing based on an artificial intelligence target detection network on the original image data, thereby obtaining target bounding box data; Step S16: performing feature extraction processing on the target bounding box data and performing target classification to obtain target type information data, wherein the target type information data includes person target data, motion target data and text target data; Step S17: performing coordinate calculation processing according to the target bounding box data and the target type information data, thereby obtaining target feature data including the target object type and position information.

3. The image transmission method for a digital camera according to claim 2, characterized in that: Step S2 includes the following steps: Step S21: performing facial feature extraction processing on the person-type target in the target feature data, thereby obtaining expression feature data; Step S22: performing emotion recognition processing on the expression feature data, when the expression feature shows a smile or a surprised expression, setting the importance weight coefficient of the character target object to 150% of the preset reference weight, otherwise the importance weight coefficient of the character target object is set to the preset reference weight, thereby obtaining the character emotion weight coefficient data; Step S23: performing motion feature extraction processing on the motion target in the target feature data, thereby obtaining target motion trajectory data; Step S24: when the target motion trajectory data determines that the target motion state is an acceleration or turning action, the importance weight coefficient of the motion target object is set to 130% of the preset reference weight, otherwise the importance weight coefficient of the motion target object is set to the preset reference weight, thereby obtaining the motion target weight coefficient data; Step S25: performing optical character recognition processing on the text-type targets in the target classification data, thereby obtaining text content data; Step S26: performing importance matching processing on the text content data based on a preset rule library, thereby obtaining text target weight coefficient data; Step S27: combining the text target weight coefficient data, the motion target weight coefficient data and the character emotion weight coefficient data into target importance data.

4. The image transmission method for a digital camera according to claim 3, characterized in that: Step S26 includes the following steps: The text content data is processed by importance matching based on the preset rule library to obtain text target weight coefficient data, wherein the preset rule library specifically sets the importance weight coefficient to 200% of the preset benchmark weight when the identified text target object is score data, and sets the importance weight coefficient to 180% of the preset benchmark weight when the identified text target object is time data; otherwise, the importance weight coefficient of the text target object is set to 120% of the preset benchmark weight.

5. The image transmission method for a digital camera according to claim 4, characterized in that: Step S3 includes the following steps: Step S31: normalizing the target importance data, mapping the weight values ​​of all target objects to a range from 0 to 1, thereby obtaining standardized weight data; Step S32: using Gaussian distribution method to perform thermal diffusion processing on the original image data based on the standardized weight data with the center point of the target object as the reference, so as to obtain single target thermal distribution data; Step S33: performing overlapping area detection processing on the single target thermal distribution data of all targets, and marking the position information of the pixel points whose thermal values ​​are greater than a preset threshold, thereby obtaining thermal overlapping area data; Step S34: performing maximum value selection processing on the thermal overlap area data to obtain thermal superposition data, wherein the maximum value selection processing selects the maximum value of all overlapping thermal values ​​as the final thermal value of the pixel point in the overlap area; Step S35: Perform global normalization processing on the thermal superposition data to obtain importance heat map data.

6. The image transmission method for a digital camera according to claim 5, characterized in that: Step S32 includes the following steps: Step S321: Calculate the thermal diffusion radius of the target object in the original image data according to the standardized weight data, so as to obtain diffusion radius data, wherein the diffusion radius is proportional to the standardized weight value, and the diffusion radius increases by 10 pixels for every increase of 0.1 in the standardized weight value; Step S322: performing two-dimensional Gaussian kernel generation processing on the diffusion radius data to obtain Gaussian diffusion template data, wherein the standard deviation of the Gaussian kernel is proportional to the diffusion radius, and the standard deviation increases by 1 unit for every 20 pixels increase in the diffusion radius; Step S323: performing target center point positioning processing on the original image data according to the position information in the target feature data, thereby obtaining heat diffusion center point data; Step S324: using the Gaussian diffusion template data to perform convolution operation on the thermal diffusion center point data, so as to obtain single target thermal distribution data.

7. The image transmission method for a digital camera according to claim 6, characterized in that: Step S33 includes the following steps: Step S331: establishing a pixel-level thermal value matrix according to the single-target thermal distribution data, and performing non-zero judgment processing on each pixel point in the matrix, thereby obtaining effective thermal area data; Step S332: extracting the region boundary of the effective thermal region data, determining the boundary contour of each target thermal distribution, and thus obtaining thermal boundary data; Step S333: Calculate the spatial relationship between the thermal regions of different targets according to the thermal boundary data, and determine whether there is an intersection between the boundaries according to the diffusion radius data, thereby obtaining the region intersection data; Step S334: performing positioning and marking processing on the pixel points in the region intersection data, and assigning a unique position index to the pixel points in each overlapping region, thereby obtaining overlapping pixel index data; Step S335: extracting the thermal value in each overlapping area according to the overlapping pixel index data, and establishing a mapping relationship between the thermal value and the pixel position, thereby obtaining thermal mapping data; Step S336: performing threshold comparison processing on the thermal value of each pixel in the thermal mapping data, screening out the pixel whose thermal value is greater than the preset threshold, thereby obtaining high thermal pixel data; Step S337: Associating the overlapping pixel index data and the high thermal pixel data to obtain thermal overlapping area data.

8. The image transmission method for a digital camera according to claim 7, characterized in that: Step S4 includes the following steps: Step S41: dividing the original image data into image blocks of equal size according to the importance heat map data, and calculating the average heat value of each image block, thereby obtaining the image block heat data; Step S42: performing hierarchical processing on the image block thermal data based on a preset first threshold and a preset second threshold, thereby obtaining hierarchical image block data, wherein the hierarchical image block data includes a high importance block, a medium importance block, and a low importance block; Step S43: performing hierarchical compression processing according to different importance levels of the hierarchical image block data, thereby obtaining high-importance compressed data, medium-importance compressed data, and low-importance compressed data; Step S44: organize and process the high-importance compressed data, the medium-importance compressed data, and the low-importance compressed data, and record the compression type and parameter information of each image block in the data header, so as to obtain layered compressed image data.

9. The image transmission method for a digital camera according to claim 8, characterized in that: Step S43 includes the following steps: Step S431: performing lossless compression encoding processing on high-importance blocks in the layered image block data, thereby obtaining high-importance compressed data; Step S432: setting JPEG compression quality parameters for the medium importance block in the layered image block data, thereby obtaining medium compression parameter data, wherein the JPEG compression quality parameter setting is specifically to linearly map the compression quality parameters according to the thermal value of the medium importance block; Step S433: performing JPEG compression processing on the medium importance block in the layered image block data according to the medium compression parameter data, thereby obtaining medium importance compressed data; Step S434: performing a quantization step index mapping process based on the average thermal value of each block on the low-importance blocks in the layered image block data, thereby obtaining low-compression parameter data, wherein the quantization step index mapping process specifically includes exponentially increasing the quantization step as the thermal value decreases; Step S435: Perform high compression ratio compression processing on low-importance blocks in the layered image block data according to the low-compression parameter data, thereby obtaining low-importance compressed data.

10. An image transmission system for a digital camera, characterized in that: Used to execute the image transmission method for a digital camera as claimed in claim 1, the image transmission system for a digital camera comprises: The image acquisition module is used to obtain the original image data taken by the digital camera; perform target scanning processing on the original image data to obtain target feature data including target object type and position information; A target weight calculation module is used to perform weight calculation processing on each target object according to the target feature data to obtain target importance data, wherein the weight calculation processing specifically includes determining the weight value based on the facial expression features of the person when the target object type is a person, determining the weight value based on the motion trajectory when the target object type is a moving target, and determining the weight value based on a preset rule when the target object type is text information; A heat map generation module is used to perform heat diffusion processing on the original image data based on the target importance data using a Gaussian distribution method with the center point of the target object as a reference to obtain importance heat map data, wherein the heat diffusion processing uses a maximum value superposition method when the heat distributions of multiple target objects overlap; The image layered compression module is used to perform layered compression processing on the original image data according to the importance heat map data to obtain layered compressed image data, wherein the layered compression processing is specifically to perform lossless compression on the area where the thermal value is greater than a preset first threshold, to perform medium compression ratio compression on the area where the thermal value is between the first threshold and the preset second threshold, and to perform high compression ratio compression on the area where the thermal value is less than the second threshold.

Images