Image optimization processing method and system based on big data

Abstract

The invention provides an image optimization processing method and system based on big data. The method comprises the following steps: acquiring an image sample containing multiple attributes from a big data set; and according to the internal relevance and semantic consistency of the image samples, classifying the plurality of image samples to form an image group with a specific topic, and automatically identifying the relevance among the plurality of image samples in combination with a deep learning model in a classification process. The continuity of the plurality of image samples in different dimensions is ensured by using a space-time consistency algorithm; analyzing the statistical characteristics of the image group, and extracting an optimization parameter set used for describing the statistical characteristics of the image group; and optimizing the plurality of image samples based on the optimization parameter set. According to the technical scheme provided by the invention, the image processing quality and efficiency are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of image processing technology, and in particular to a method and system for image optimization processing based on big data. Background Art

[0002] With the development of the internet and multimedia technologies, image data has been widely used in social media, online education, virtual reality, augmented reality, and other fields. These applications require processing not only static images but also a variety of image data types, such as dynamic image sequences and 3D reconstruction models. To meet these demands, image optimization technologies must be able to handle large-scale and diverse datasets while maintaining image quality while improving visual quality, compression efficiency, and user satisfaction.

[0003] Several image optimization solutions are currently available on the market. These typically involve the following steps: first, obtaining image samples from a dataset; second, performing simple classification or labeling on these samples; and finally, applying uniform optimization parameters to the classified images. These solutions primarily rely on traditional image processing techniques, such as histogram equalization and sharpening filters, and typically use fixed optimization parameters for all images.

[0004] However, existing image optimization processing solutions have the following shortcomings: Lack of personalized processing: Existing image optimization methods often use uniform optimization parameters to process all images, ignoring the differences between different types of images and the unique needs of image groups for specific themes. This approach can lead to poor optimization results, especially when processing complex and diverse image data, making it difficult to meet the high-quality requirements of different application scenarios. Low classification accuracy: Traditional image classification methods often rely on simple feature matching or manual labeling. Such classification methods are susceptible to noise interference and have difficulty capturing the deep correlation between images. In addition, they do not fully consider the coherence of images in the temporal and spatial dimensions, which may lead to inaccurate classification results, thus affecting subsequent optimization effects. Fixed optimization parameters: Optimization parameters in existing solutions are typically pre-set and cannot be dynamically adjusted based on the specific content of the image. This means that even images within the same category may require different optimization processes due to varying content. Fixed parameter methods cannot adapt to this situation, thus limiting the flexibility and effectiveness of the optimization. Summary of the Invention

[0005] The embodiments of the present application provide a method and system for image optimization processing based on big data, which are used to solve the problems in the prior art such as lack of personalized processing, low classification accuracy, and fixed optimization parameters.

[0006] In a first aspect, an embodiment of the present application provides an image optimization processing method based on big data, comprising: Acquire image samples containing multiple attributes from a large data set, wherein the image samples include static pictures, dynamic image sequences, and 3D reconstructed models; Classifying the plurality of image samples based on their inherent relevance and semantic consistency to form image groups with specific themes, wherein the classification process automatically identifies the relevance between the plurality of image samples in conjunction with a deep learning model, and utilizes a spatiotemporal consistency algorithm to ensure the coherence of the plurality of image samples in different dimensions; Analyzing statistical characteristics of the image group and extracting an optimized parameter set for describing the statistical characteristics of the group images, the optimized parameter set includes a color space transformation matrix, a texture enhancement factor, a structural integrity preservation coefficient, and a dynamic content adaptation coefficient; Optimizing the plurality of image samples based on the optimization parameter set.

[0007] In a second aspect, an embodiment of the present application provides an image optimization processing system based on big data, comprising: An acquisition module is used to acquire image samples containing multiple attributes from a large data set, wherein the image samples include static pictures, dynamic image sequences, and 3D reconstructed models; a classification module, configured to classify the plurality of image samples according to their intrinsic relevance and semantic consistency to form image groups with specific themes, wherein the classification process automatically identifies the relevance between the plurality of image samples in conjunction with a deep learning model, and utilizes a spatiotemporal consistency algorithm to ensure the coherence of the plurality of image samples in different dimensions; an extraction module configured to analyze statistical characteristics of the image group and extract an optimized parameter set for describing the statistical characteristics of the group images, the optimized parameter set comprising a color space transformation matrix, a texture enhancement factor, a structural integrity preservation coefficient, and a dynamic content adaptation coefficient; An optimization module is used to optimize the plurality of image samples based on the optimization parameter set.

[0008] In an embodiment of the present application, image samples containing multiple attributes are obtained from a large data set, and the image samples include static pictures, dynamic image sequences, and 3D reconstruction models; based on the intrinsic correlation and semantic consistency of the image samples, multiple image samples are classified to form image groups with specific themes, and the classification process is combined with a deep learning model to automatically identify the correlation between multiple image samples, and a spatiotemporal consistency algorithm is used to ensure the coherence of multiple image samples in different dimensions; by analyzing the statistical characteristics of the image group, and extracting an optimized parameter set for describing the statistical characteristics of the group image, the optimized parameter set includes a color space transformation matrix, a texture enhancement factor, a structural integrity preservation coefficient, and a dynamic content adaptation coefficient; based on the optimized parameter set, multiple image samples are optimized. The technical solution provided by the present application improves the quality and efficiency of image processing.

[0009] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0011] Figure 1 A flowchart of an image optimization processing method based on big data provided in an embodiment of the present application; Figure 2 A comparison chart of the image before and after optimization processing provided in the embodiment of the present application; Figure 3 A schematic diagram of the structure of an image optimization processing system based on big data provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0013] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0014] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0015] Figure 1 A flowchart of an image optimization processing method based on big data is provided for the embodiment of the present application. Figure 1 As shown, the method includes: 101. Acquire image samples containing multiple attributes from a large data set, wherein the image samples include static images, dynamic image sequences, and 3D reconstructed models; The concepts involved in this step mainly include: Large datasets: refers to a collection of large amounts of image data, which usually comes from different channels and is diverse and complex. Image samples: refers to image instances selected from a large data set for subsequent processing, including static images, dynamic image sequences (DIS), and 3D reconstruction models; Static Image: refers to a single-frame image. Common formats include JPEG, PNG, BMP, etc. Dynamic image sequence: refers to a video stream consisting of a series of continuous image frames, which can be in standard video formats such as MP4, AVI, etc. 3D reconstruction model: refers to a 3D model synthesized from images taken from multiple angles using computer vision technology. It is commonly used in fields such as virtual reality (VR) and augmented reality (AR).

[0016] In this embodiment of the application, it is assumed that there is a large dataset containing image data from various sources such as social media, surveillance cameras, user-uploaded photos and videos, and 3D models generated by game engines. To demonstrate how to obtain image samples from such a large dataset, an example is provided: Static images: User-uploaded images captured from social media platforms. These images may be JPEG-formatted landscapes, portraits, etc. Dynamic image sequence: continuous image frames captured from the video recorded by the surveillance camera, in MP4 format; 3D reconstruction model: Game scene model exported from a popular 3D game engine in FBX format.

[0017] The specific implementation steps are as follows: Data collection: Use a crawler to capture user-uploaded images from social media platforms (such as Weibo and Instagram) and save them in JPEG format. Also, obtain a video from a surveillance camera system and extract several frames as image sequence samples. Data preprocessing: Preprocess the collected image data, including removing duplicate images, cropping useless edges, and adjusting the size to ensure data consistency and availability; 3D model export: Use the game development software Unity to export a scene in the game as a 3D model and save it in FBX format; Data integration: The three types of image samples mentioned above are integrated into the same database so that subsequent processing steps can easily access and operate these data.

[0018] Specific numerical examples in the embodiment: Suppose 1,000 JPEG images are captured from a social media platform, each with a size of 1024x768 pixels; a 1-minute video is obtained from a surveillance camera, with a resolution of 1280x720 pixels and a frame rate of 30fps, from which 1,800 frames are extracted as image sequence samples; finally, a 3D model containing 500 polygons is exported from Unity.

[0019] 102. Classify the plurality of image samples according to the intrinsic relevance and semantic consistency of the image samples to form an image group with a specific theme; The classification process combines a deep learning model to automatically identify the correlation between the multiple image samples, and uses a spatiotemporal consistency algorithm to ensure the coherence of the multiple image samples in different dimensions. Optionally, in step 102, "the step of classifying the plurality of image samples based on the intrinsic correlation and semantic consistency of the image samples to form an image group with a specific theme" includes: applying a graph theory algorithm to construct an image connection network between the plurality of image samples, wherein each node in the connection network represents an image sample, and the weights of the edges in the connection network are determined by the similarity between the plurality of image samples, the continuity and contextual consistency of the plurality of image samples in the time dimension, and the semantic labels of the plurality of image samples; and identifying a compact subset in the image connection network to form an image group.

[0020] The concepts involved in this step mainly include: Image samples: refers to image samples with multiple attributes obtained from large data sets, including static images, dynamic image sequences, and 3D reconstructed models; Intrinsic relevance: refers to the similarity between image samples, including visual similarity (such as color, texture, shape, etc.) and content relevance (such as the same event, place or theme); Semantic consistency: refers to the consistency of image samples at the semantic level, that is, the consistency of the theme or content expressed by the image; Graph theory algorithms are mathematical tools used to study the properties and structure of graphs and can be used to construct a network of connections between images. In this step, graph theory algorithms are used to build a relationship graph between image samples. Image Link Network: This is a network structure that represents image samples as nodes and the similarity relationship between images as edges. Each node represents an image sample, and the weight of the edge represents the similarity between images. Tight subset: refers to a subgraph with a high density of edges between nodes in an image connection network, that is, a set of images with close connections, which can be used to identify image groups; Deep learning model: A machine learning technology that can automatically learn features from data and use them for tasks such as image classification and recognition. Common deep learning models include Convolutional Neural Network (CNN); Spatiotemporal consistency algorithm: refers to an algorithm that ensures the consistency of images in the temporal dimension when processing dynamic image sequences or time series data. This can ensure the coherence of image samples in different dimensions; Contextual consistency: refers to the similarity of image samples in content and context, ensuring the relevance between image samples.

[0021] In this embodiment of the application, it is assumed that image samples with various attributes have been obtained from a large data set, including static images, dynamic image sequences, and 3D reconstruction models. Next, these image samples need to be classified into groups of images with specific themes. The specific implementation steps are as follows: Apply deep learning models to automatically identify correlations between image samples: Use pre-trained convolutional neural network (CNN) to extract features from image samples; Based on the extracted feature vectors, the similarity between image samples is calculated.

[0022] Use spatiotemporal consistency algorithms to ensure the coherence of image samples in different dimensions: For dynamic image sequences, ensure that the similarity between adjacent frames is high enough.

[0023] For static images and 3D reconstructed models, semantic tags or other metadata are used to assist in determining whether they belong to the same subject.

[0024] Build an image connectivity network and identify dense subsets: Represent each image sample as a node in the network; The weight of the edge represents the similarity between image samples, and the continuity and context consistency in the time dimension as well as the semantic label are used to determine the weight of the edge; Use community discovery algorithms (such as the Louvain algorithm) to identify compact subsets to form groups of images.

[0025] In a specific embodiment, it is assumed that the following image samples have been obtained from a large data set: 1000 static images; A 1-minute video (30fps) was used to extract 1800 frames as image sequence samples; 50 3D reconstruction models.

[0026] Feature extraction: Use a pre-trained ResNet-50 model to extract features from each still image, each video frame, and each keyframe of the 3D reconstruction model. Assume that the feature vector dimension of each image is 2048.

[0027] Similarity calculation: For any two image samples and Calculate the cosine similarity (CosineSimilarity) between their feature vectors: ; in and They are image samples and The eigenvector of .

[0028] Coherence of dynamic image sequences: For video image sequences, calculate the similarity between adjacent frames and ensure that the similarity between adjacent frames is higher than a certain threshold (e.g. 0.7); The weight of the edge between adjacent frames can be expressed as: ; This ensures that the weight between adjacent frames does not fall below 0.7.

[0029] Coherence between static images and 3D models: Use OCR technology to identify text content in static images and match it with relevant descriptions of 3D models to ensure semantic consistency.

[0030] For the similarity between static images and 3D models, the weight of the edge can be determined by semantic label matching. Assuming that the similarity between a picture containing "beach" and a 3D model labeled "beach" is 0.9, the weight of the edge is 0.9.

[0031] Consistency of semantic tags: For static images and 3D models, label matching is used to determine the weight of the edge. For example, if two static images both contain the label "beach", the weight of the edge between them can be set to 0.8.

[0032] Build the network: Represent each image sample as a node in the network; If the similarity between two image samples is greater than 0.5, an edge is added to the network and assigned a corresponding weight.

[0033] Identify tight subsets: Use the Louvain algorithm to identify dense subsets in an image connectivity network. Suppose three dense subsets are identified, each containing 400 static images, 900 frames of dynamic image sequences, and 10 3D models.

[0034] Example values: Assume that the following values ​​are calculated: The dimension of the feature vector extracted by the ResNet-50 model is 2048; Calculate the cosine similarity between static images, for example, two static images and The similarity between them is 0.85; The similarity between adjacent frames in a video image sequence is 0.75; The similarity between static images and 3D models is matched by semantic labels. For example, the similarity between a picture containing "beach" and a 3D model labeled "beach" is 0.9.

[0035] Specific implementation example: Construct an image connection network: Assume there are 1000 static images, 1800 frames of video image sequence and 50 3D models; Use the ResNet-50 model to extract feature vectors; Calculate the similarity between each pair of image samples.

[0036] Determine the edge weights: The weight of the edge between adjacent frames is 0.75; The weight of the edge between the static image and the 3D model is 0.9; The weight of the edges between static images is 0.85.

[0037] Identify tight subsets: Use Louvain algorithm to identify tight subsets; Suppose three compact subsets are identified, containing 400 static images, 900 frames of dynamic image sequences, and 10 3D models respectively.

[0038] 103. Analyze the statistical characteristics of the image group and extract an optimized parameter set for describing the statistical characteristics of the group images; The optimization parameter set includes a color space transformation matrix, a texture enhancement factor, a structural integrity preservation coefficient, and a dynamic content adaptation coefficient; Optionally, the "analyzing the statistical characteristics of the image group and extracting an optimized parameter set for describing the statistical characteristics of the group image" in step 103 includes: performing statistical analysis of the color histogram, texture features and structural features of the multiple image samples in each of the image groups to obtain statistical results; determining the corresponding main eigenvectors based on the statistical results, and generating an optimized parameter set based on the main eigenvectors.

[0039] The concepts involved in this step mainly include: Statistical characteristics: refers to the statistical data describing image characteristics, such as color distribution, texture characteristics, structural characteristics, etc., extracted by analyzing image samples within an image group; Principal Feature Vector: The feature vector that best represents the characteristics of the image group extracted through statistical analysis can be used to generate the optimization parameter set; Optimization parameter set: refers to a set of parameters used to optimize image processing, including the Color Space Transformation Matrix (CSTM), Texture Enhancement Factor (TEF), Structural Integrity Retention Coefficient (SIRC), and Dynamic Content Adaptation Coefficient (DCAC); Color Histogram: describes the distribution of different colors in the image; Texture Features: Describes the details of the texture in the image, such as texture contrast, uniformity, etc. Structural Features: describe the edge and structure information in the image; Color Space Transformation Matrix (CSTM): used to convert the color space of an image to improve its color performance; Texture Enhancement Factor (TEF): used to enhance the texture details in the image to make it clearer; Structural Integrity Retention Coefficient (SIRC): used to maintain the basic structure of the image and prevent the loss of important structural information during the optimization process; Dynamic Content Adaptation Coefficient (DCAC): used to adapt to content changes in dynamic image sequences and ensure consistency between different frames.

[0040] In this embodiment, it is assumed that the image samples have been classified into several image groups with specific themes using graph theory algorithms and community discovery algorithms. Now, it is necessary to further analyze the statistical characteristics of these image groups and extract the optimized parameter set accordingly. The specific implementation steps are as follows: Statistical property analysis: Color histogram: Calculate the color histogram of the images in each image group; Texture features: extract texture features of images in each image group; Structural features: Analyze the structural features of images within each image group; Extract the optimized parameter set: According to the statistical results, the corresponding main eigenvectors are determined, and the optimization parameter set is generated according to the main eigenvectors.

[0041] In a specific embodiment, it is assumed that the image samples have been classified into three compact subsets (image groups), each of which contains different types of image samples. Now, statistical characteristics analysis will be performed on these image groups to extract the optimized parameter set.

[0042] Color histogram: Calculate the color histogram for each image sample in each image group. The color histogram describes the distribution of different colors in the image; For example, for the 400 static images in the first group, calculate the color histogram of each image and summarize the color distribution at the group level; Texture features: Extract texture features of images within each image group, such as Gray-Level Co-occurrence Matrix (GLCM) features; For example, the gray-level co-occurrence matrix (GLCM) is used to extract texture features and calculate the texture feature statistics at the group level.

[0043] Structural features: Analyze the structural features of images in each image group, such as edge detection results, contour information, etc. For example, edge detection algorithms (such as Canny edge detection) are used to extract edge information and calculate group-level structural feature statistics.

[0044] Determine the dominant eigenvector: Use principal component analysis (PCA) or other dimensionality reduction techniques to identify the most important eigenvectors from the statistical results; For example, through PCA analysis, the principal component vector of the color histogram of the first group is obtained as , the principal component vector of texture features is , the principal component vector of structural features is .

[0045] Generate an optimized parameter set: Generate corresponding optimization parameter sets based on main eigenvectors for subsequent image optimization processing; For example, for the first group, the color histogram principal component vector Generate color space transformation matrix , according to the texture feature principal component vector Generate texture enhancement factor , according to the principal component vector of structural characteristics Generate Structural Integrity Retention Factor .

[0046] Example values: Assume that after statistical analysis of the 400 static images in the first image group, the following values ​​are obtained: Color histogram: The color distribution at the group level is red accounting for 20%, green accounting for 30%, and blue accounting for 50%; Texture features: The statistical values ​​of texture features at the group level are GLCM contrast of 0.4, energy of 0.3, and homogeneity of 0.7; Structural characteristics: The average edge strength at the group level is 0.15; Through PCA analysis, the principal component vector of the color histogram is obtained is [0.3, 0.4, 0.3], the principal component vector of texture features is [0.5, 0.3, 0.2], the principal component vector of the structural feature is [0.2, 0.6, 0.2].

[0047] Based on these principal component vectors, the optimized parameter set can be generated: Color space transformation matrix :according to The resulting matrix can be ; Texture Enhancement Factor :according to The generated factor can be 0.5; Structural integrity retention factor :according to The generated coefficient can be 0.6;

[0048] In this way, an optimized parameter set is generated for each image group, so that the images can be optimized according to these parameters in the subsequent steps.

[0049] Summary of the complete implementation: Statistical property analysis: Calculate the color histograms of the 400 static images in the first group and summarize the color distribution at the group level; Extract texture features of the 400 static images in the first group and calculate group-level texture feature statistics; The structural features of the 400 static images in the first group were analyzed, and the group-level structural feature statistics were calculated.

[0050] Extract the optimized parameter set: Use PCA analysis to get the principal component vector of the color histogram is [0.3, 0.4, 0.3], the principal component vector of texture features is [0.5, 0.3, 0.2], the principal component vector of the structural feature is [0.2, 0.6, 0.2].

[0051] Generating an Optimized Parameter Set: Color Space Transformation Matrix Texture enhancement factor , structural integrity retention factor .

[0052] 104. Optimize the plurality of image samples based on the optimization parameter set; In the embodiment of this application, it is assumed that the image samples have been classified and the optimization parameter set has been extracted based on the statistical characteristics of each image group. Now, the image samples need to be optimized based on these optimization parameter sets. The specific implementation steps are as follows: Apply a color space transformation matrix: Convert the color space of the image according to the color space transformation matrix to improve the color performance of the image; Apply a texture enhancement factor: Enhance the texture details in the image according to the texture enhancement factor to make it clearer; Apply the structural integrity retention factor: The basic structure of the image is maintained according to the structural integrity preservation coefficient to prevent the loss of important structural information during the optimization process.

[0053] Apply dynamic content adaptation coefficient (applicable only to dynamic image sequences): Adapts to content changes in dynamic image sequences based on the dynamic content adaptation coefficient to ensure consistency between different frames.

[0054] In this specific embodiment, it is assumed that an optimized parameter set has been extracted for the first image group (including 400 static images), and the following values ​​have been obtained: Color space transformation matrix for ; Texture enhancement factor is 0.5; Structural integrity retention factor is 0.6.

[0055] Color space conversion: For each still image in the first group, apply the color space transformation matrix Perform color space conversion; Assume that the RGB color channel value of a picture is , then the new color channel value for: ; For example, for an image with RGB values ​​of [100, 150, 50], apply The new RGB value is: ; Texture Enhancement: For each still image in the first group, apply a texture enhancement factor Perform texture enhancement; Assume that the texture feature vector of a picture is , then the enhanced texture feature vector for: ; For example, for a picture with a texture feature vector of [0.4, 0.3, 0.2], apply The new texture feature vector is: ; Structure Retention: For each static image in the first group, apply the structural integrity preservation factor Maintain the basic structure of the image; Assume that the structural feature vector of a picture is , then the structural feature vector after preservation for: ; For example, for a picture with a structural feature vector of [0.2, 0.6, 0.2], apply The new structural feature vector is: ; Summary of the complete implementation: Color space conversion: Apply the color space transformation matrix to each still image in the first group Perform color space conversion; For example, for an image with RGB values ​​of [100, 150, 50], apply The new RGB value after is [100, 105, 100].

[0056] Texture Enhancement: Apply a texture enhancement factor to each still image in the first group Perform texture enhancement; For example, for a picture with a texture feature vector of [0.4, 0.3, 0.2], apply The new texture feature vector is [0.2, 0.15, 0.1].

[0057] Structure Retention: Apply a structural integrity preservation factor to each still image in the first group Maintain the basic structure of the image; For example, for a picture with a structural feature vector of [0.2, 0.6, 0.2], apply = 0.6, the new structural feature vector is [0.12, 0.36, 0.12].

[0058] Optionally, it also includes: the extraction of the optimization parameter set is based on a comprehensive consideration of the common features of multiple image samples in the image group, the differential features between multiple image samples, and the time series change trends of multiple image samples to ensure the consistency of the optimization results.

[0059] The concepts involved in this step mainly include: Common Features: refers to the features shared by multiple image samples in an image group, such as color distribution, texture type, etc. Distinctive Features: refers to the distinguishing features between different image samples in an image group, such as specific differences in color histograms, differences in texture details, etc. Temporal Change Trend: refers to the trend of image samples changing over time in a dynamic image sequence, such as the change pattern between adjacent frames; Consistency: During the optimization process, it ensures that the optimization results of all image samples in the image group are consistent and coherent.

[0060] In this embodiment of the application, it is assumed that the image samples have been classified and the optimization parameter set has been extracted based on the statistical characteristics of each image group. Now, it is necessary to ensure the consistency of the optimization results based on a comprehensive consideration of the common features, differential features, and time series change trends of multiple image samples within the image group. The specific implementation steps are as follows: Extract common features: Perform statistical analysis on color histograms, texture features, and structural features of multiple image samples within an image group to extract common features; Extract differential features: Analyze the difference features between different image samples in the image group and consider how to distinguish and process these difference features during the optimization process; Analyze time series change trends (applicable to dynamic image sequences): Perform time series analysis on image samples in dynamic image sequences, extract time series change trends, and ensure that the change patterns between adjacent frames are maintained during the optimization process; Comprehensive consideration to generate optimized parameter set: Combining common features, differential features and time series change trends, an optimization parameter set is generated to ensure the consistency of the optimization results.

[0061] In this specific embodiment, assuming that image samples have been classified and statistical characteristics of the first image group (including 400 static images) have been obtained, it is now necessary to further extract common features, differential features, and time series change trends, and generate an optimized parameter set.

[0062] Color histogram analysis: Calculate the color histogram of the 400 static images in the first group and summarize the color distribution at the group level; Assume that the color distribution at the group level is red accounting for 20%, green accounting for 30%, and blue accounting for 50%.

[0063] Texture feature analysis: The texture features of the 400 static images in the first group are extracted, and the texture feature statistics at the group level are calculated.

[0064] Assume that the statistical values ​​of texture features at the group level are GLCM contrast 0.4, energy 0.3, and homogeneity 0.7.

[0065] Structural feature analysis: The structural features of the 400 static images in the first group were analyzed, and the group-level structural feature statistics were calculated.

[0066] Assume that the average edge strength at the group level is 0.15.

[0067] Color histogram differences: Compare the color histograms of each static image in the first group to find samples with greater differences; Suppose that some images have more than 25% red, less than 28% green, and less than 48% blue.

[0068] Texture feature differences: Compare the texture features of each static image in the first group and find samples with greater differences; Suppose some images have GLCM contrast exceeding 0.45, energy below 0.25, and homogeneity below 0.65.

[0069] Differences in structural characteristics: Compare the structural features of each static image in the first group to find samples with greater differences; Suppose that the average edge strength of some images exceeds 0.2.

[0070] Time series trend: Perform time series analysis on image samples in dynamic image sequences and extract time series change trends; Assume that the color between adjacent frames in a dynamic image sequence does not change much, but the texture and structure change greatly; For example, the color histogram similarity between adjacent frames is 0.8, the texture feature similarity is 0.6, and the structural feature similarity is 0.7.

[0071] Comprehensive consideration to generate optimized parameter set Generate an optimized parameter set by combining common features, differential features and time series change trends; For example, for the first group, the generated optimized parameter set may include: Color space transformation matrix ; Texture enhancement factor ; Structural integrity retention factor ; Dynamic content adaptation coefficient (Applicable only to dynamic image sequences).

[0072] Example values: Assume the following analysis is performed on the first group of images: Color histogram: The color distribution at the group level is red (20%), green (30%), and blue (50%). Texture features: The statistical values ​​of texture features at the group level are GLCM contrast of 0.4, energy of 0.3, and homogeneity of 0.7; Structural characteristics: The average edge strength at the group level is 0.15; Differentiation features: In some images, the red ratio exceeds 25%, the green ratio is less than 28%, and the blue ratio is less than 48%; in some images, the GLCM contrast exceeds 0.45, the energy is less than 0.25, and the homogeneity is less than 0.65; in some images, the average edge strength exceeds 0.2; Time series change trend (dynamic image sequence): The color histogram similarity between adjacent frames is 0.8, the texture feature similarity is 0.6, and the structural feature similarity is 0.7.

[0073] Optionally, the method further comprises: generating an optimization parameter set according to the main feature vector: Define a multidimensional fitness function; Iteratively optimizing the multidimensional adaptation function through a genetic algorithm or a reinforcement learning mechanism to find an optimal parameter set; The multidimensional adaptation function is defined as follows: ; in, Represents a multidimensional fitness function used to evaluate the optimization parameter set Adaptability to specific goals, Indicates the use of optimized parameter set Visual quality scores of optimized image groups, Indicates the use of optimized parameter set Compression efficiency score of the optimized image group, Indicates the use of optimized parameter set Optimized user preference score, coefficient Represent the importance weights of visual quality score, compression efficiency score and user preference score over time t, and satisfy .

[0074] The concepts involved in this step mainly include: Multidimensional Fitness Function: used to evaluate the optimization parameter set Adaptability to specific goals. By defining a multi-dimensional adaptation function, multiple optimization goals can be comprehensively considered, such as visual quality, compression efficiency, and user preferences; Genetic Algorithm (GA): A search algorithm that simulates the evolutionary process in nature and searches for the optimal solution through operations such as selection, crossover, and mutation. Reinforcement Learning Mechanism (RL): A machine learning method that learns which actions to take to maximize cumulative rewards through the interaction of an agent with its environment. Visual Quality Score ): Evaluate using the optimized parameter set Optimized visual quality of image groups; Compression Efficiency Score ): Evaluate using the optimized parameter set Optimized compression efficiency of image groups; User Preference Score ): Evaluate using the optimized parameter set User preference for the optimized image groups; coefficient : represents the importance weights of visual quality score, compression efficiency score and user preference score as time t changes, and satisfies .

[0075] In this embodiment, it is assumed that the image samples have been classified and the optimized parameter set has been extracted based on the statistical characteristics of each image group. Now, it is necessary to iteratively optimize the multidimensional adaptation function through a genetic algorithm or reinforcement learning mechanism to find the optimal parameter set. The specific implementation steps are as follows: Define the multidimensional fitness function: ,in represents the set of optimization parameters; Iteratively optimize the multidimensional fitness function through genetic algorithms or reinforcement learning mechanisms: Using genetic algorithms or reinforcement learning mechanisms, the multidimensional fitness function is iteratively optimized to find the optimal parameter set.

[0076] In a specific embodiment, it is assumed that an optimized parameter set has been extracted for a first image group (including 400 static images) and a multi-dimensional adaptation function has been defined. Now, it is necessary to optimize the multi-dimensional adaptation function using a genetic algorithm to find the optimal parameter set.

[0077] Initialize the population: Generate an initial population, each individual represents a set of optimized parameters ; Assuming the initial population size is 100, each individual consists of four parameters: color space transformation matrix , texture enhancement factor , Structural integrity retention factor and dynamic content adaptation coefficients DC AC.

[0078] Calculate fitness: Calculate the multidimensional fitness function for each individual ; Assumptions ,but .

[0079] Select an action: Select individuals with higher fitness to enter the next generation; Assume that the top 50% of individuals are selected as parents.

[0080] Crossover operation: Perform crossover operation on the selected parent individuals to generate new individuals; Assume the crossover probability is 0.7.

[0081] Mutation operation: Perform mutation operations on the newly generated individuals to introduce random disturbances; Assume that the mutation probability is 0.1.

[0082] Iterative optimization: Repeat the selection, crossover, and mutation operations until a predetermined number of iterations or convergence conditions are reached; Example values: Assume the following initial population and calculation results: The initial population size is 100; Each individual consists of four parameters: color space transformation matrix , texture enhancement factor , Structural integrity retention factor and dynamic content adaptation coefficients DC AC; Assume that the initial parameter range is: .

[0083] Multidimensional fitness function ; Assume that the parameters of an individual are ; Assume that the fitness of the individual is ; Through genetic algorithms, the multi-dimensional fitness function is iteratively optimized to eventually find the optimal parameter set. Assume that after 100 iterations, the optimal parameter set is found: The optimal parameter set is ; The optimal fitness is .

[0084] Optionally, it also includes: the visual quality score Calculated as follows: ; in, is the original image group, Is to use the optimized parameter set Optimized image groups, represents the L2 norm, which is the sum of the squared Euclidean distances between two image groups; represents the L1 norm, which is the sum of the absolute distances between two image groups; represents the L0 norm, which refers to the number of non-zero elements between two image groups; N is the number of test image groups; Indicates that for all test image groups Perform summation; is the standard deviation over time t; and Expressed as the equilibrium constant; represents the average score of the global difference before and after optimization for all test image groups; Represents the sum of the differences before and after optimization for all test image groups; represents the sum of the differences in non-zero pixels before and after optimization for all test image groups.

[0085] In this step, the visual quality score is used to evaluate the optimization parameter set An indicator of image optimization effectiveness. This indicator comprehensively considers the global difference, local difference, and non-zero pixel difference between the image before and after optimization.

[0086] In the embodiment of the present application, it is assumed that there is a set of test image groups N=100, each of which contains an original image and an optimized image The goal is to evaluate the visual quality score of the optimized image relative to the original image.

[0087] Example; L2 norm part: Suppose in a certain test, So for a specific image , the sum of the squares of the pixel differences before and after optimization is 10000, then: ; After calculation, , so the average score for this section is about 0.135.

[0088] L1 norm part: ; LO norm part: Assuming that for the same image The number of different pixels before and after optimization is 100, and the total number of pixels in the image is 1000, then: ; Final visual quality score: Assuming equilibrium constant and ,but: ; Optionally, it also includes: the compression efficiency score Calculated as follows: ; in, Indicates the Original image groups The file size, Indicates the use of optimized parameter set Optimized image groups The file size of N is the number of test images, Indicates the ratio of the difference in file size of the image group before and after optimization to the size of the original image group. Indicates that for all test image groups Perform the summation, Represents the average percentage difference in file size before and after optimization for all tested image groups; The concepts involved in this step mainly include: Compression Efficiency Score ): Evaluate using the optimized parameter set The compression efficiency of the optimized image group relative to the original image group. The compression efficiency score is measured by calculating the ratio of the change in file size of the image group before and after optimization.

[0089] File Size: refers to the storage space occupied by the image file; The ratio of the file size difference before and after optimization: refers to the ratio of the difference between the optimized image file size and the original image file size to the original image file size.

[0090] In the embodiment of the present application, it is assumed that a multi-dimensional adaptation function has been defined and the compression efficiency score needs to be calculated. .

[0091] Now, we will demonstrate how to calculate the compression efficiency score through a specific numerical example.

[0092] In a specific embodiment, suppose there is an image group containing 400 static images, and the difference in file size before and after optimization needs to be calculated and the compression efficiency score is used. To evaluate the optimization effect.

[0093] Calculate the original file size: For the i-th original image group , calculate its file size Size( ).

[0094] Assume there are 5 image groups, each containing 80 images, and the file sizes are shown in Table 1 below: Table 1 ; Apply the optimized parameter set: Using optimized parameter sets Optimize image groups; Calculate the optimized image group file size Size( ); Assume that the optimized file size is as shown in Table 2 below: Table 2 ; Calculate the ratio: For each image group Calculate the ratio of the file size difference before and after optimization .

[0095] Assume that the calculation results are shown in Table 3 below: Table 3 ; Calculate the average ratio: Calculate the average percentage difference in file size before and after optimization for all test image groups; Assume the calculation results are as follows: ; Therefore, the compression efficiency score is 0.10206, indicating that on average, using the optimized parameter set The file size of the optimized image group is reduced by approximately 10.206%.

[0096] Summary of the complete implementation: Collect the original image group file size: Calculate the original file size of each image group; Assume that the original file sizes are 5MB, 4.5MB, 6MB, 4MB, and 5.5MB respectively.

[0097] Calculate the file size of the optimized image group: Applying optimized parameter sets Optimize image groups; Assume that the optimized file sizes are 4.5MB, 4MB, 5.5MB, 3.5MB, and 5MB respectively.

[0098] Calculate the ratio of the file size difference before and after optimization: For each image group, calculate the ratio of the file size difference before and after optimization; Assume that the difference proportions are 0.1, 0.1111, 0.0833, 0.125, and 0.0909 respectively.

[0099] Calculate the compression efficiency score: Calculate the average percentage difference in file size before and after optimization for all test image groups.

[0100] Assume a compression efficiency score of 0.10206.

[0101] The user preference score Calculated as follows: ; in, Represents the optimized parameter set for the j-th user preference, is the weight of the corresponding user category that changes with time t, and is the adjustment coefficient, M is the total number of categories preferred by users, is the average preference of all users, represents the preference at the previous moment, represents the weighted sum of preferences of different user groups. Indicates the relative difference between the preferences of a specific user group and the average preferences of all user groups. Focus on the temporal dynamics of user preferences.

[0102] In this step, the user preference score Used to measure the optimization parameter set The impact on the preferences of different types of users. This score consists of three parts: the weighted sum of the preferences of user groups, the relative difference between the preferences of a specific user group and the average preferences of all user groups, and the time dynamics of user preferences.

[0103] In this embodiment of the present application, it is assumed that there are three user categories M=3, each category has its own preference score The goal is to calculate the User preference ratings under .

[0104] Example: Weighted sum of user group preferences: Assume there are three user categories, and the preference of each category at a certain point in time is , The corresponding weight .

[0105] ; Calculate the logarithmic part: ; therefore: ; The relative difference between the preferences of a specific user group and the average preferences of all user groups: Assume that the average preference of all users is ;

[0106] ; Calculate the logarithmic part: ; therefore: ; Time dynamics of user preferences: Assume that the preference at the previous moment ;

[0107] ; Calculate the logarithmic part: ; therefore: ; Final User Preference Rating: Hypothetical Adjustment Factor and ,but: ; Furthermore, the visual quality score, compression efficiency score, and user preference score calculated above may be substituted into a multidimensional adaptation function, and the multidimensional adaptation function may be iteratively optimized through a genetic algorithm or a reinforcement learning mechanism to find an optimal parameter set.

[0108] The following is a specific example for steps 101 to 104. Figure 2 As shown: Based on the image optimization processing method of the present application, the system first extracts the original landscape image (2.5MB) with low contrast (0.3), blurred texture (0.2), and loose structure (0.4) from a large data set, and classifies it into the "natural landscape" theme group through a deep learning model and a spatiotemporal consistency algorithm; then, the common characteristics of the group (such as the ubiquitous low brightness histogram peak and weak texture variance) are analyzed to automatically generate an optimization parameter set: the color space transformation matrix is ​​used to increase the contrast to 0.85, the texture enhancement factor is used to enhance the detail clarity to 0.92, and the structural integrity preservation coefficient is used to sharpen the edge to 0.88; finally, based on the iterative optimization of the multidimensional adaptation function (balancing visual quality, compression efficiency and user preference), the image file size is reduced by 28% (to 1.8MB), while the visual quality score jumps from 3.2 to 8.7, achieving a dual improvement in image quality and efficiency.

[0109] Figure 3 The present invention provides a schematic diagram of a structure of an image optimization processing system based on big data. Figure 3 As shown, the device includes: An acquisition module 21 is configured to acquire image samples containing multiple attributes from a large data set, wherein the image samples include static images, dynamic image sequences, and 3D reconstructed models; A classification module 22 is configured to classify the plurality of image samples based on their intrinsic relevance and semantic consistency to form image groups with specific themes, wherein the classification process automatically identifies the relevance between the plurality of image samples in conjunction with a deep learning model and utilizes a spatiotemporal consistency algorithm to ensure the coherence of the plurality of image samples in different dimensions; an extraction module 23 for analyzing the statistical characteristics of the image group and extracting an optimization parameter set for describing the statistical characteristics of the image group, the optimization parameter set including a color space transformation matrix, a texture enhancement factor, a structural integrity preservation coefficient, and a dynamic content adaptation coefficient; The optimization module 24 is configured to optimize the plurality of image samples based on the optimization parameter set.

[0110] Figure 3 The image optimization processing system based on big data can be executed Figure 1 The implementation principle and technical effects of the big data-based image optimization processing method described in the illustrated embodiment will not be elaborated on here. The specific manner in which each module and unit performs operations in the big data-based image optimization processing system in the above embodiment has been described in detail in the embodiment of the method and will not be elaborated on here.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for image optimization processing based on big data, characterized in that: include: Acquire image samples containing multiple attributes from a large data set, wherein the image samples include static pictures, dynamic image sequences, and 3D reconstructed models; Classifying the plurality of image samples based on their inherent relevance and semantic consistency to form image groups with specific themes, wherein the classification process automatically identifies the relevance between the plurality of image samples in conjunction with a deep learning model, and utilizes a spatiotemporal consistency algorithm to ensure the coherence of the plurality of image samples in different dimensions; Analyzing statistical characteristics of the image group and extracting an optimized parameter set for describing the statistical characteristics of the group, the optimized parameter set includes a color space transformation matrix, a texture enhancement factor, a structural integrity preservation coefficient, and a dynamic content adaptation coefficient; Optimizing the plurality of image samples based on the optimization parameter set.

2. The method according to claim 1, characterized in that The extraction of the optimization parameter set is based on a comprehensive consideration of the common features of the multiple image samples in the image group, the difference features between the multiple image samples, and the time series change trend of the multiple image samples to ensure the consistency of the optimization results.

3. The method according to claim 1, characterized in that The method of classifying the plurality of image samples according to the intrinsic relevance and semantic consistency of the image samples to form an image group with a specific theme includes: Applying a graph theory algorithm to construct an image connection network between the plurality of image samples, wherein each node in the connection network represents one of the image samples, and the weights of the edges in the connection network are determined by the similarity between the plurality of image samples, the continuity and contextual consistency of the plurality of image samples in the time dimension, and the semantic labels of the plurality of image samples; Tight subsets of the image connectivity network are identified to form image groups.

4. The method according to claim 3, characterized in that The step of analyzing the statistical characteristics of the image group and extracting an optimized parameter set for describing the statistical characteristics of the group images includes: Performing statistical analysis on color histograms, texture features, and structural features of the plurality of image samples in each of the image groups to obtain statistical results; According to the statistical results, the corresponding main eigenvectors are determined, and an optimization parameter set is generated according to the main eigenvectors.

5. The method according to claim 4, characterized in that Generating an optimization parameter set according to the main feature vector includes: Define a multidimensional fitness function; Iteratively optimizing the multidimensional adaptation function through a genetic algorithm or a reinforcement learning mechanism to find an optimal parameter set; The multidimensional adaptation function is defined as follows: ； in, Represents a multidimensional fitness function used to evaluate the optimization parameter set Adaptability to specific goals, Indicates the use of optimized parameter set Visual quality scores of optimized image groups, Indicates the use of optimized parameter set Compression efficiency score of the optimized image group, Indicates the use of optimized parameter set Optimized user preference score, coefficient Represent the importance weights of visual quality score, compression efficiency score and user preference score over time t, and satisfy .

6. The method according to claim 5, characterized in that The visual quality score Calculated as follows: ； in, is the original image group, Is to use the optimized parameter set Optimized image groups, represents the L2 norm, which is the sum of the squared Euclidean distances between two image groups; represents the L1 norm, which is the sum of the absolute distances between two image groups; represents the L0 norm, which refers to the number of non-zero elements between two image groups; N is the number of test image groups; represents the sum of all test image groups i; is the standard deviation over time t; and Expressed as the equilibrium constant; represents the average score of the global difference before and after optimization for all test image groups; Represents the sum of the differences before and after optimization for all test image groups; represents the sum of the differences in non-zero pixels before and after optimization for all test image groups.

7. The method according to claim 5, characterized in that The compression efficiency score Calculated as follows: ； in, Indicates the Original image groups The file size, Indicates the use of optimized parameter set Optimized image groups The file size of N is the number of test images, Indicates the ratio of the difference in file size of the image group before and after optimization to the size of the original image group. Indicates that for all test image groups Perform the summation, Represents the average percentage difference in file size before and after optimization for all tested image groups.

8. The method according to claim 5, characterized in that The user preference score Calculated as follows: ； in, Indicates the Class user pair optimization parameter set preference, is the weight of the corresponding user category that changes with time t, and is the adjustment coefficient, M is the total number of categories preferred by users, The average preference of all users, represents the preference at the previous moment, represents the weighted sum of preferences of different user groups. Indicates the relative difference between the preferences of a specific user group and the average preferences of all user groups. It focuses on the temporal dynamic changes of user preferences.

9. An image optimization processing system based on big data, characterized in that: include: An acquisition module is used to acquire image samples containing multiple attributes from a large data set, wherein the image samples include static pictures, dynamic image sequences, and 3D reconstructed models; a classification module, configured to classify the plurality of image samples according to their intrinsic relevance and semantic consistency to form image groups with specific themes, wherein the classification process automatically identifies the relevance between the plurality of image samples in conjunction with a deep learning model, and utilizes a spatiotemporal consistency algorithm to ensure the coherence of the plurality of image samples in different dimensions; an extraction module, configured to analyze statistical characteristics of the image group and extract an optimized parameter set for describing the statistical characteristics of the group, the optimized parameter set comprising a color space transformation matrix, a texture enhancement factor, a structural integrity preservation coefficient, and a dynamic content adaptation coefficient; An optimization module is used to optimize the plurality of image samples based on the optimization parameter set.

Images