Picture identifying and cutting system and method based on artificial intelligence

Through the image recognition and cropping system based on artificial intelligence, using MTCNN model and gradient direction histogram and other technologies, the image cropping and adjustment are automated, which solves the complex and low efficiency of the image cropping process in traditional technology, and achieves efficient and accurate pictures Processing and business demand response.

CN119027678APending Publication Date: 2024-11-26XIAN SINGULARITY ENERGY TECH CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202410986942.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

In the development of business-oriented systems, the traditional image cutting and compression process requires the collaboration of multiple technical jobs, and the lack of unified technical standards leads to low tooling and intelligence, long delivery cycle, low quality and compatibility, and the inability to respond to business quickly need.

Method used

Adopt an artificial intelligence-based image recognition and cropping system, including significant area detection module, picture input module, preprocessing module, crop processing module, picture control module and cache module, and use the MTCNN model, gradient direction histogram and support vector machine for picture cropping and adjust to achieve automated processing.

Benefits of technology

Through machine learning and modular parameterization methods, images are automatically processed, manual intervention is reduced, development efficiency is improved, business customization needs are supported, and the accuracy and efficiency of image cropping is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure BDA0004957686890000071
    Figure BDA0004957686890000071
  • Figure BDA0004957686890000081
    Figure BDA0004957686890000081
  • Figure BDA0004957686890000091
    Figure BDA0004957686890000091
Patent Text Reader

Abstract

The invention belongs to the technical field of picture processing, and particularly discloses a picture recognition cutting system and method based on artificial intelligence, and the system comprises a salient region detection module which is used for outputting a two-dimensional matrix containing a key feature information contour; a picture input module; the preprocessing module is used for receiving the two-dimensional matrix and adjusting pixel values based on the two-dimensional matrix and original image pixels; the cutting processing module is used for cutting the original picture based on the adjusted pixel value to obtain a standby picture data set; the picture control module is used for processing the standby picture data set based on specific requirements of a user to obtain a target picture data set; the caching module is used for caching the target picture data set; and the user interface module provides an access interface for a user and is used for acquiring the target picture data set. According to the invention, the picture content is identified when the picture is cut according to different user requirements, and the cut picture is ensured to contain the significant features required by the user.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to an image recognition and cropping system and method based on artificial intelligence. Background Art

[0002] Currently, many companies face the need to access image resources of different sizes and qualities in multiple scenarios, including PC and mobile terminals (including APP / mini-programs / H5, etc.) during the process of business system development. The traditional development model involves cropping and compressing image attachments into different images when uploading them, front-end and back-end development and joint debugging, and reading images of different sizes and compression ratios according to different business requirements during testing. The entire process requires participation and collaboration of many personnel and technical trades (front-end engineers, back-end file storage services), and the entire process often lacks unified technical standards (such as cropping algorithms for images with different aspect ratios, compression ratios, etc.). The degree of tooling and intelligence is low, and the collaboration efficiency between various parts is low, resulting in long delivery cycles, low quality and compatibility, and an inability to respond quickly and efficiently to business and customer needs. Summary of the Invention

[0003] The purpose of the present invention is to overcome the defects in the prior art and provide an image recognition and cropping system based on artificial intelligence.

[0004] A first aspect of the present invention provides an artificial intelligence-based image recognition and cropping system, comprising:

[0005] Salient region detection module: used to detect and mark salient regions in the original image and output a two-dimensional matrix containing the outline of key feature information;

[0006] Image input module: connected to the salient area detection module, used to send the original image to the salient area detection module;

[0007] Preprocessing module: connected to the salient area detection module, receiving the two-dimensional matrix of the salient area detection module, and adjusting the pixel values ​​based on the two-dimensional matrix and the original image pixels;

[0008] Cropping processing module: connected to the pre-processing module, crops the original image based on the adjusted pixel values ​​to obtain a backup image data set;

[0009] Image control module: connected to the cropping processing module, processing the backup image data set based on user-specific requirements to obtain a target image data set;

[0010] Cache module: connected to the image control module, used to cache the target image data set;

[0011] User interface module: provides an access interface for users to obtain the target image dataset.

[0012] A further solution is that the salient area detection module includes a key feature information detector, which performs pre-detection on the original image based on the gradient direction histogram and the support vector machine to obtain a pre-detection image with a boundary cropping box, and then optimizes the pre-detection image through the MTCNN model to complete the marking of the key feature information of the pre-detection image and obtain a two-dimensional matrix containing the outline of the key feature information.

[0013] A further solution is that the MTCNN model includes three networks: Proposal Network, RefineNetwork and Output Network. The Proposal Network is used for candidate box generation, the Refine Network is used for candidate box optimization, and the Output Network is used for key point detection.

[0014] The training process of the MTCNN model is staged training. Each network is trained separately and then combined for joint training. The training process of the MTCNN model is as follows:

[0015] The Proposal Network receives an original image and outputs the location of the candidate box and the confidence level of whether it contains key feature information;

[0016] The Refine Network receives the candidate boxes and their features extracted by the Proposal Network, and outputs more accurate candidate boxes and their confidence levels after screening and adjustment.

[0017] The Output Network receives the candidate box output by the Refine Network and outputs the coordinates of the candidate box, the location of the key feature information, and the attributes of the key feature information.

[0018] A further solution is that the MTCNN model is also used to identify specific demand features. Based on the user's specific needs, a picture set containing several specific demand features is input into the MTCNN model, and the coordinates of the specific demand features are output.

[0019] A further solution is that the process of the image input module sending the original image to the salient area detection module is:

[0020] Determine whether the original image is an image file accessible to the salient area detection module. If so, send the original image directly to the salient area detection module. If not, download the original image through the PATH information.

[0021] Determine whether the format of the original image received by the salient area detection module is supported. If so, the salient area detection module identifies and marks the original image. If not, throw an exception and return the original image.

[0022] After verification, the image decoding library is used to decode the image into the image data structure of the image control module, and all color spaces of the original image are converted into RGB color space;

[0023] Correct the image direction and determine whether the image contains rotation information. If so, correct the image direction based on the rotation information.

[0024] A further solution is that the picture data structure is:

[0025] Serial number property Data Type meaning 1 w Integer Image width after cropping 2 h Integer Image cropping height 3 q Integer The quality of the image after processing, the value is a percentage 4 r Integer Rotation angle 5 fv Flag Flip Horizontal 6 fh Flag Flip vertically 7 cx Integer Use coordinate clipping, starting pixel coordinates 8 cw Integer Use coordinate clipping and custom width 9 ch Integer Use coordinate clipping and custom height 10 x Integer Scaling

[0026] A further solution is that the process of pre-detecting the original image using the gradient direction histogram and the support vector machine is as follows:

[0027] Load the original image and convert it into a grayscale image;

[0028] Initialize the gradient direction histogram and support vector machine;

[0029] Detect key feature information in the grayscale image. The key feature information detector returns a bounding box list representing the detected key feature information area. If the bounding box list is empty, it means that there is no key feature information in the image. If the bounding box list is not empty, the bounding cropping box is drawn according to the bounding box list.

[0030] A second aspect of the present invention provides an artificial intelligence-based image recognition and cropping method, which uses the above-mentioned image recognition and cropping system and includes the following steps:

[0031] Step 1: Input the original image information to the salient area detection module through the image input module, and receive the original image based on the file stream method;

[0032] Step 2: Use the salient region detection module to detect and mark salient regions in the original image, and obtain a two-dimensional matrix containing the outline of key feature information;

[0033] Step 3: Use the preprocessing module to adjust the pixel values ​​based on the two-dimensional matrix information and the original image pixels;

[0034] Step 4: The image is cropped based on the pre-processed information through the cropping processing module;

[0035] Step 5: The image control module further processes the image based on specific requirements and adjusts it to the required image data structure;

[0036] Step 6: Cache the processed specific requirement images into the cache system through the cache module;

[0037] Step 7: Provide users with an access interface through the user interface module, complete image operations through the interface, and obtain processed results.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] The present invention avoids repeated manual recognition, image processing, code writing, interface documentation, and testing through machine learning, modularization, and parameterization. It identifies image content when cropping images according to different user needs, ensuring that the cropped images contain the significant features required by the user.

[0040] The present invention uses a training set to generate image attachments with unified rules, realizing full automation from manual configuration to machine recognition operation. Each module is independent of each other, and the entire service supports dynamic replacement and expansion through high-availability components, which greatly improves the production efficiency of development and testing personnel. The generation availability of automated code can be further improved by componentizing the front-end and back-end. By reserving code extension points, developers can expand and maintain more easily, supporting the customization needs of the business. This method is highly innovative and practical in actual business system development. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The following drawings are merely provided for illustrative purposes only and are not intended to limit the scope of the present invention.

[0042] Figure 1 : Schematic diagram of the connection of image recognition and cropping system modules;

[0043] Figure 2 : Flowchart of the picture input process;

[0044] Figure 3 : Original image pre-detection flow chart;

[0045] Figure 4 : Flowchart of the image recognition and cropping method of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution, design method and advantages of the present invention more clear, the present invention is further described in detail below through specific embodiments in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0047] Example 1

[0048] like Figure 1As shown, the present invention provides an artificial intelligence-based image recognition and cropping system, comprising:

[0049] The salient region detection module detects and labels salient regions in the original image, outputting a two-dimensional matrix containing the outlines of key feature information. Key feature information refers to the portions of the image that are not desired to be cut off or lost after cropping. The rows of the two-dimensional matrix represent the pixel positions from top to bottom, while the columns represent the pixel positions from left to right.

[0050] Image input module: connected to the salient area detection module, used to send the original image to the salient area detection module;

[0051] Preprocessing module: connected to the salient region detection module, receiving the two-dimensional matrix from the salient region detection module, and adjusting the pixel values ​​based on the two-dimensional matrix and the original image pixels. The adjustment process is as follows: first, the width and height of the image are scaled using a bilinear interpolation algorithm. This algorithm calculates the value of the new pixel by weighted averaging of adjacent pixels to maintain a smooth image transition; then, functions provided by an image processing library (such as the PIL library) are used to perform format conversion using built-in codecs, such as converting PNG format to JPEG format; and then, the image quality is adjusted using the JPEG compression algorithm to reduce the file size.

[0052] Cropping processing module: connected to the pre-processing module, crops the original image based on the adjusted pixel values ​​to obtain a backup image data set;

[0053] Image control module: connected to the cropping processing module, processing the backup image data set based on user-specific requirements to obtain a target image data set;

[0054] Cache module: connected to the image control module, used to cache the target image data set;

[0055] User interface module: provides an access interface for users to obtain the target image dataset.

[0056] In the above, the salient area detection module includes a key feature information detector, which performs pre-detection on the original image based on the gradient direction histogram and the support vector machine to obtain a pre-detection image with a boundary cropping frame, and then optimizes the pre-detection image through the MTCNN model to complete the marking of the key feature information of the pre-detection image and obtain a two-dimensional matrix containing the outline of the key feature information. In this embodiment, the key feature information can be any image information such as face information, cabinet information, box information, etc. Among them, if Figure 3As shown, the process of the key feature information detector performing pre-detection on the original image based on the gradient direction histogram and the support vector machine is as follows: loading the original image and converting the original image into a grayscale image; initializing the gradient direction histogram and the support vector machine; detecting the key feature information in the grayscale image, and the key feature information detector returns a bounding box list representing the detected key feature information area. If the bounding box list is empty, it means that there is no key feature information in the image. If the bounding box list is not empty, a bounding cropping box is drawn according to the bounding box list.

[0057] In the above, the MTCNN model includes three networks: Proposal Network, Refine Network and Output Network. The Proposal Network is used for candidate box generation, the Refine Network is used for candidate box optimization, and the Output Network is used for key point detection. The training process of the MTCNN model is staged training. Each network is trained separately and then combined for joint training. The training process of the MTCNN model is as follows:

[0058] The Proposal Network receives an original image and outputs the location of the candidate box and the confidence level of whether it contains key feature information;

[0059] The Refine Network receives the candidate boxes and their features extracted by the Proposal Network, and outputs more accurate candidate boxes and their confidence levels after screening and adjustment.

[0060] The Output Network receives the candidate box output by the Refine Network and outputs the coordinates of the candidate box, the location of the key feature information, and the attributes of the key feature information.

[0061] In this embodiment, the MTCNN model can not only identify key features such as faces, but also identify specific features based on user needs. For example, different companies may have different specific cropping requirements, meaning they want different key features to be retained after cropping. Therefore, a set of images containing several specific features can be input into the MTCNN model, which then outputs the coordinates of those specific features.

[0062] Before training the MTCNN model, a large number of original images should be divided into a training set and a validation set. After the MTCNN model training is completed, the validation set needs to be used to verify and optimize the model. Specifically, the validation indicators include accuracy, recall rate, precision, F1 score, ROC curve and AUC value. The specific meanings of the validation indicators are shown in the table below:

[0063]

[0064] The optimization process of the MTCNN model is:

[0065] Accuracy optimization:

[0066] Increase or decrease the number of model network layers to improve the model's expressiveness; use larger datasets for training to enhance the model's generalization capabilities; apply data augmentation techniques, such as rotation, cropping, and flipping, to increase data diversity;

[0067] Recall rate optimization:

[0068] Adjust the loss function to focus more on recall; adjust the threshold to reduce the probability of detecting false negatives; use more positive samples for training to improve the model's ability to identify positive samples;

[0069] Accuracy optimization:

[0070] Adjust the loss function to focus more on accuracy; adjust the threshold to reduce the probability of detecting false positives; apply more complex post-processing techniques to reduce false positives;

[0071] F1 score optimization:

[0072] Comprehensively optimize precision and recall to achieve a balance between the two; adjust model architecture and hyperparameters to improve overall model performance; apply cross-validation techniques to select the optimal model parameters;

[0073] ROC curve and AUC value optimization:

[0074] Use more complex model architectures to improve the ability to distinguish between positive and negative samples; perform hyperparameter tuning to select optimal model parameters; apply higher-quality datasets to reduce noise and bias.

[0075] like Figure 2 As shown, in the above, the process of the image input module sending the original image to the salient area detection module is:

[0076] Determine whether the original image is an image file accessible to the salient area detection module. If so, send the original image directly to the salient area detection module. If not, download the original image through the PATH information.

[0077] Determine whether the format of the original image received by the salient area detection module is supported. If so, the salient area detection module identifies and marks the original image. If not, throw an exception and return the original image.

[0078] After verification, the image decoding library is used to decode the image into the image data structure of the image control module, and all color spaces of the original image are converted into RGB color space;

[0079] Correct the image direction and determine whether the image contains rotation information. If so, correct the image direction based on the rotation information.

[0080] The picture data structure is:

[0081]

[0082]

[0083] Example 2

[0084] The scenario of this embodiment is to use the technology stack of Django+OpenCV+PIL+dlib+TensorFlow+PyTorch+awssdk+jackson to implement an artificial intelligence-based image content recognition and cropping method, combined with the recognition and cropping system of Example 1. When performing image recognition and cropping, Figure 4 As shown, the following steps are included:

[0085] Step 1: Use Django or other web service technology stacks to provide a set of long link services for the model to receive images, supporting multiple image formats (JPEG, PNG, GIF, etc.).

[0086] Step 2: Provide a set of model cli programs through Linux Shell to receive images in the computer command line interactive window.

[0087] Step 3: After the model receives the original image information, it uses OpenCV or PIL library to perform image preprocessing:

[0088] First, determine whether it is an image file accessible to the local machine. Otherwise, download the original image through the PATH information and verify whether the format in the image is supported. Otherwise, throw an exception and return the original image. After the verification is passed, it is convenient for subsequent processing. First, use libjpeg, libpng, libtiff and other image decoding libraries to decode the image. This step decodes the image into an image data structure in the computer memory. Unify the color space of the image. Because images of different image formats may use different color spaces (for example: RGB, YUV, CMYK), in order to facilitate processing, all color spaces of the image are converted into RGB color space for the next step of processing. Correct the image direction first. For example, in a jpeg format image, EXIF ​​metadata is included. It is necessary to determine whether it contains the image rotation information (orientation). If it contains rotation information, correct the image direction according to this rotation information.

[0089] Step 4: Use the dlib library to implement the gradient orientation histogram (HOG) and support vector machine (SVM) for key feature information detection:

[0090] Load the original image and convert it to grayscale for ease of subsequent processing. Use the key feature detector to detect key features in the grayscale image. The key feature detector returns a bounding box list representing the detected key feature areas. If the bounding box list is empty, it indicates that there is no key feature information in the image and no further key feature recognition and extraction is required. If the bounding box list is not empty, a bounding cropping box is drawn based on the bounding box list.

[0091] Step 5: Use TensorFlow or PyTorch to implement MTCNN and complete salient area detection.

[0092] The premise of step 5 is that the model has been trained using the training dataset. The training steps are:

[0093] Collect an image dataset containing the target object. For salient region detection, you can use a public image dataset or a custom dataset (for example, a company selling apples could use a dataset of different apples to test the model's ability to identify apples as the primary content of the image). Label the salient regions and key objects in the image. Typically, a labeling tool is used to generate bounding boxes.

[0094] Preprocess images and annotation data, such as normalization, data augmentation (rotation, flipping, cropping, etc.);

[0095] Construct the MTCNN model. The MTCNN model consists of three stages: P-Net (Proposal Network), R-Net (Refine Network) and O-Net (Output Network), which are used for candidate box generation, candidate box optimization and key point detection respectively.

[0096] Training the MTCNN model needs to be carried out in stages, with each network trained separately and then combined for joint training.

[0097] P-Net (Proposal Network) receives an original image and outputs the location of the candidate box and the confidence level of whether it contains key feature information;

[0098] R-Net (Refine Network) receives the candidate boxes and their features extracted by P-Net, and outputs more accurate candidate boxes and their confidence levels after screening and adjustment;

[0099] O-Net (Output Network) receives the candidate box and its features output by R-Net, and outputs the coordinates of the candidate box, the location of key feature information, and the attributes of key feature information (product front logo, product model logo, etc.).

[0100] Use the validation set to evaluate the performance of the model, including indicators such as accuracy and recall, and adjust the model parameters.

[0101] If the training is complete, the key feature recognition steps are:

[0102] Load the image to be processed and convert it to the appropriate format;

[0103] Use MTCNN to detect key feature information in the image;

[0104] Recursively obtain salient area information in the detected result set;

[0105] Crop the original image based on the acquired salient area information;

[0106] Step 6: After receiving the output of the previous step, the image processing module first verifies whether the model parameters already exist in the training set. The built-in parameters of this application are:

[0107] Serial number property Data Type Example meaning 1 w Integer 50 Image width after cropping 2 h Integer 50 Image cropping height 3 q Integer 30 The quality of the image after processing, the value is a percentage (1 to 100) 4 r Integer 90 Rotation angle 5 fv Flag [flag] Flip Horizontal 6 fh Flag [flag] Flip vertically 7 cx Integer 50 Use coordinate clipping, starting pixel coordinates 8 cw Integer 50 Use coordinate clipping and custom width 9 ch Integer 50 Use coordinate clipping and custom height 10 x Integer 100 Scaling

[0108] Step 7: Check whether there is a cache in the cache module (memory, disk, OSS). If there is a cache, directly return the information in the cache and the process ends. Otherwise, continue with the subsequent process.

[0109] Step 8: According to the lexical analysis, the image attachment address is obtained and the actual image is obtained. Because this step is a network asynchronous scenario, additional exception handling is required (network exception, processing exception, other exceptions, exception capture, and exception compensation).

[0110] Step 9: Adjust, crop, filter, and encode the processed image to form a new image.

[0111] Step 10: Put the processed image in response and stdout and return it to the user side and terminal.

[0112] Step 11: Cache the processed image in the cache system (memory, disk, OSS).

[0113] While various embodiments of the present invention have been described above, the above descriptions are intended to be illustrative, non-exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or technological improvements in the marketplace, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. An artificial intelligence-based image recognition and cropping system, characterized in that: include: Salient region detection module: used to detect and mark salient regions in the original image, and output a two-dimensional matrix containing the outline of key feature information; Image input module: connected to the salient area detection module, used to send the original image to the salient area detection module; Preprocessing module: connected to the salient area detection module, receiving the two-dimensional matrix output by the salient area detection module, and adjusting the pixel values ​​based on the two-dimensional matrix and the original image pixels; Cropping processing module: connected to the pre-processing module, performs cropping processing on the original image based on the adjusted pixel values ​​to obtain a spare image data set; Image control module: connected to the cropping processing module, processing the spare image data set based on user-specific requirements to obtain a target image data set; Cache module: connected to the image control module, used to cache the target image data set; User interface module: provides an access interface for users to obtain the target image dataset.

2. The image recognition and cropping system based on artificial intelligence according to claim 1, characterized in that: The salient area detection module includes a key feature information detector, which performs pre-detection on the original image based on the gradient direction histogram and the support vector machine to obtain a pre-detection image with a boundary cropping box, and then optimizes the pre-detection image through the MTCNN model to complete the marking of the key feature information of the pre-detection image, and obtains a two-dimensional matrix containing the outline of the key feature information.

3. The image recognition and cropping system based on artificial intelligence according to claim 2, characterized in that: The MTCNN model includes three networks: Proposal Network, Refine Network and Output Network. The Proposal Network is used for candidate box generation, the Refine Network is used for candidate box optimization, and the Output Network is used for key point detection. The training process of the MTCNN model is staged training. Each network is trained separately and then combined for joint training. The training process of the MTCNN model is as follows: ProposalNetwork receives an original image and outputs the location of the candidate box and the confidence of whether it contains key feature information; RefineNetwork receives the candidate boxes and their features extracted by ProposalNetwork, and outputs more accurate candidate boxes and their confidence after screening and adjustment; OutputNetwork receives the candidate box output by RefineNetwork and outputs the coordinates of the candidate box, the location of key feature information, and the attributes of key feature information.

4. The image recognition and cropping system based on artificial intelligence according to claim 3, characterized in that: The MTCNN model is also used to identify specific demand features. Based on user specific needs, a picture set containing several specific demand features is input into the MTCNN model, and the coordinates of the specific demand features are output.

5. The image recognition and cropping system based on artificial intelligence according to claim 1, characterized in that: The process of the picture input module sending the original picture to the salient area detection module is as follows: Determine whether the original image is an image file accessible to the salient area detection module. If so, directly send the original image to the salient area detection module. If not, download the original image through PATH information. Determine whether the format of the original image received by the salient area detection module is supported. If so, identify and mark the original image through the salient area detection module. If not, throw an exception and return the original image. After verification, the original image is decoded into the image data structure of the image control module using the image decoding library, and all color spaces of the original image are converted into RGB color space; Correct the image direction and determine whether it contains image rotation information. If so, correct the image direction based on the rotation information and send it to the salient area detection module.

6. The artificial intelligence-based image recognition and cropping system according to claim 5, characterized in that: The picture data structure is:

7. The artificial intelligence-based image recognition and cropping system according to claim 2, characterized in that: The process of the key feature information detector pre-detecting the original image based on the gradient direction histogram and the support vector machine is as follows: Load the original image and convert it into a grayscale image; Initialize the gradient direction histogram and support vector machine; Key feature information is detected in the grayscale image. The key feature information detector returns a bounding box list representing the detected key feature information area. If the bounding box list is empty, it means that there is no key feature information in the image. If the bounding box list is not empty, a bounding cropping box is drawn according to the bounding box list to obtain a pre-detection image with a bounding cropping box.

8. An image recognition and cropping method based on artificial intelligence, characterized in that: The artificial intelligence-based image recognition and cropping system according to any one of claims 1 to 7 is applied, comprising the following steps: Step 1: Input the original image information to the salient area detection module through the image input module, and receive the original image based on the file stream method; Step 2: Detect and mark the salient areas of the original image through the salient area detection module to obtain a two-dimensional matrix containing the outline of key feature information; Step 3: The pixel values ​​are adjusted according to the two-dimensional matrix information and the original image pixels through the preprocessing module; Step 4: The image is cropped based on the preprocessed information through the cropping processing module; Step 5: The image control module further processes the image based on specific requirements and adjusts it to a required image data structure; Step 6: Cache the processed specific requirement images to the cache system through the cache module; Step 7: Provide users with an access interface through the user interface module, complete image operations through the interface, and obtain processed results.

Citation Information

Patent Citations

  • Cropping method and device for people pictures

    CN104484854A

  • Intelligent picture cutting method and system, computer equipment and storage medium

    CN111524145A

  • Rotating target detection method and device based on Gaussian distribution

    CN115424267A

  • Text detection method and system for traditional Chinese printed old newspaper

    CN117746447A