Computer vision recognition system and method

Abstract

The present invention discloses a computer vision recognition system and method, and specifically relates to the technical field of image data recognition, and is used to solve the problem of inaccurate machine vision recognition. The present invention first acquires sample data, acquires image information in different dimensions through visual recognition, and performs feature extraction on the acquired image information, uses a local binary pattern to compare pixel points in the image with neighborhood pixels, determines the generated binary pattern, and then reduces the dimension, uses a generative adversarial network to generate image tasks according to feature vectors, optimizes the generated image according to a generator and a discriminator, determines the process of computer-generated art images and performs detection, and finally analyzes various data generated in the process of computer-generated art images to determine the accuracy of visual recognition, thereby improving the accuracy of recognition in the process of computer-generated art and reducing the waste of computing resources.

Description

Technical Field

[0001] The present invention relates to the field of image data recognition technology, and more specifically, to a computer vision recognition system and method. Background Art

[0002] Computer vision recognition system is a system that uses computer technology to simulate and implement human visual perception. These systems usually use image or video data as input and use algorithms and models to identify, analyze and understand the information in the image.

[0003] Computer-generated art refers to the process of generating artworks using computer algorithms, programs and data. This art form combines computer science, mathematics, graphics and creative arts, making computers a tool and medium for artistic creation. Computer-generated art includes static images, animations, music, literary works and other digital art forms.

[0004] Deficiencies of existing technologies:

[0005] In the process of computer-generated art, the generation process combines multiple disciplines and large amounts of data. In the context of multidisciplinary intersection, generated artworks have more complex effects. Due to the great diversity of computer-generated art, in different styles, forms and themes, computer vision models have unstable recognition performance when faced with such diversity, and there are difficulties in integrating and processing multi-source information, which leads to the inability to accurately generate the expected digital art, affecting the creative quality and expressiveness of computer-generated art, and causing unnecessary waste of computing resources. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a computer vision recognition system and method, which first obtains image information in different dimensions through visual recognition, and performs feature extraction on the obtained image information, uses a local binary pattern to compare the pixel points in the image with the neighborhood pixels, determines the generated binary pattern, and then reduces the dimension, uses a generative adversarial network to generate image tasks according to the feature vector, optimizes the generated image according to the generator and the discriminator, determines the process of computer-generated art images and performs detection, and finally analyzes the various data generated in the process of computer-generated art images to determine the accuracy of visual recognition, so as to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The computer vision recognition method comprises the following steps:

[0009] Acquire sample data, obtain image information in different dimensions through visual recognition, and perform feature extraction on the acquired image information;

[0010] Image processing and pattern recognition are performed during the feature extraction process, and the dimension of the image features is reduced. The pixels in the image are compared with the neighboring pixels using local binary patterns to determine the generated binary pattern;

[0011] Extract local binary patterns of pixels according to binary patterns to form feature vectors, use generative adversarial networks to generate image tasks according to feature vectors, optimize generated images according to generators and discriminators, determine the process of computer-generated art images and perform detection;

[0012] Analyze various data generated in the process of computer-generated artistic images, determine the accuracy of visual recognition, and manage visual recognition based on the results.

[0013] In a preferred embodiment, a local binary pattern is used to compare a pixel in an image with a neighboring pixel to determine a generated binary pattern. The specific process is as follows:

[0014] Scan the entire image, and for each pixel in the image, select the neighborhood around it;

[0015] Compare the grayscale value of the central pixel with the grayscale value of each pixel in the neighborhood to obtain a binary sequence;

[0016] The obtained binary sequence is converted into a binary number as the local binary pattern of the pixel.

[0017] In a preferred embodiment, the local binary pattern of the pixel point is extracted according to the binary pattern to form a feature vector, and the specific process is as follows:

[0018] Use the statistical histogram of LBP mode to represent image information;

[0019] Collect image samples and use the LBP operator to extract the LBP features of each image. The LBP features of all images are combined into a matrix, where each row represents the LBP feature vector of an image.

[0020] Normalize the LBP feature matrix by the mean, and subtract the mean of the corresponding column from the value of each column;

[0021] Calculate the covariance matrix of the LBP feature matrix, and perform eigenvalue decomposition on the covariance matrix to obtain the corresponding eigenvalues ​​and corresponding eigenvectors, and use the eigenvectors to form a new feature space, where each eigenvector corresponds to a principal component;

[0022] The original LBP feature matrix is ​​projected into the low-dimensional space composed of the selected principal components for dimensionality reduction.

[0023] In a preferred embodiment, a generative adversarial network is used to generate image tasks according to feature vectors, and the generated images are optimized according to the generator and the discriminator. The specific process is as follows:

[0024] A random noise vector is input through the generator of the generative adversarial network, upsampled using the deconvolution layer, and the noise vector is mapped to a result similar to the real result;

[0025] The generator of the generative adversarial network downsamples the generated result through the convolution layer, maps the generated result into a binary classification result and outputs it.

[0026] In a preferred embodiment, the data generated in the process of computer-generated artistic images are analyzed to determine the accuracy of visual recognition. The specific process is as follows:

[0027] Analyze various data generated in the process of computer-generated art images to obtain style determination information and tuning influence information in the process of computer-generated art images;

[0028] The style determination information includes the structural quality similarity index, and the tuning impact information includes the expression stability index;

[0029] The obtained structural quality similarity index and expression stability index are calculated to obtain the recognition generation accuracy coefficient;

[0030] The structural quality similarity index and expression stability index are proportional to the recognition generation accuracy coefficient;

[0031] The resulting recognition generation accuracy coefficient is compared to the recognition generation evaluation threshold.

[0032] In a preferred embodiment, the generated recognition generation accuracy coefficient is compared with the recognition generation evaluation threshold, and the specific process is as follows:

[0033] Compare the recognition generation accuracy coefficient with the recognition generation evaluation threshold;

[0034] If the recognition generation accuracy coefficient is greater than or equal to the recognition generation evaluation threshold, a recognition generation normal signal is generated to maintain normal management;

[0035] If the recognition generation accuracy coefficient is less than the recognition generation evaluation threshold, an recognition generation fluctuation signal is generated and regulation and management are performed.

[0036] A computer vision recognition system, used in the above-mentioned computer vision recognition method, comprising:

[0037] The sample preprocessing module acquires sample data, obtains image information in different dimensions through visual recognition, extracts features from the acquired image information, compares pixels in the image with neighboring pixels using local binary patterns, determines the generated binary pattern, and reduces the dimension;

[0038] The recognition and analysis module uses a generative adversarial network to generate image tasks based on feature vectors, optimizes the generated images based on the generator and discriminator, determines the process of computer-generated art images and performs detection;

[0039] The management and control module analyzes various data generated in the process of computer-generated artistic images, determines the accuracy of visual recognition, and manages visual recognition based on the results.

[0040] Technical effects and advantages of the present invention:

[0041] The present invention first acquires sample data, acquires image information in different dimensions through visual recognition, performs feature extraction on the acquired image information, uses local binary patterns to compare pixels in the image with neighborhood pixels, determines the generated binary pattern, and then reduces the dimension, uses a generative adversarial network to generate image tasks according to feature vectors, optimizes the generated image according to the generator and the discriminator, determines the process of computer-generated art images and performs detection, and finally analyzes various data generated in the process of computer-generated art images to determine the accuracy of visual recognition, thereby improving the accuracy of recognition in the process of computer-generated art and reducing the waste of computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The figure is a flow chart of the computer vision recognition method of the present invention.

[0043] Figure 2 It is a structural schematic diagram of the computer vision recognition system of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] Example 1

[0046] like Figure 1 As shown, a computer vision recognition method, the method comprising:

[0047] The visual identity system is a unified use of basic identification elements such as logos, fonts and colors to create a unique and consistent visual expression system for companies, brands and individuals, including the following elements;

[0048] Logo: The brand logo is the core of brand identification, usually a graphic, symbol or text, representing the brand's unique identity and image. The design elements of the logo should be consistently applied throughout the brand to ensure the unity of the brand image;

[0049] Fonts: Font selection plays a key role in brand visual identity. A consistent font style helps convey the brand's professionalism, style, and personality. The consistent use of fonts allows the brand to maintain a consistent visual style across different media and occasions.

[0050] Typography and layout: The typography and layout in brand documents, advertisements and other design elements are also important components of visual identity. A consistent typographic style helps enhance the professionalism and overall feel of the brand.

[0051] Color: Color is an important part of brand identity. Different colors convey different emotions and images. By consistently using specific colors in logos, advertisements, packaging, etc., brands can establish a unique emotional connection in the minds of consumers.

[0052] Graphic elements and patterns: In addition to the logo, some specific graphic elements or patterns may also become part of the brand's visual identity, which can enhance the brand's recognition and play a unique visual guiding role in the design;

[0053] By uniformly applying basic elements, the visual identity system conveys a consistent brand image across different media and channels, thereby establishing a strong and consistent brand impression among consumers.

[0054] Dynamic visual recognition mainly involves processing video data, that is, a series of continuous image frames, not just a single static image. It is suitable for scenes that require real-time monitoring and analysis of continuous changes, such as video surveillance, motion analysis, behavior recognition, etc. Through the understanding of the time dimension, it can capture and analyze the movement, changes, and dynamic scenes in dynamic streaming media (video, animated images). In the process of computer art generation, the feature extraction of elements in dynamic streaming media affects the quality of digital art. Using dynamic visual recognition to obtain features includes:

[0055] Feature extraction in dynamic visual recognition can better capture the key information of the image and improve the accuracy of subsequent classification or recognition. For example, texture, color and shape features are commonly used feature types in the field of image processing and can effectively characterize the structure and information in the image. The three types of features are as follows:

[0056] Texture features: Texture is an important structural feature in an image, reflecting the spatial relationship between pixels. Texture features describe the repetition or change pattern of an image region. In dynamic visual recognition, texture features can be used to distinguish objects of different materials or surfaces, as well as to capture texture information in motion;

[0057] Color features: Color is an important factor in human visual perception and is also crucial for image recognition. Color features describe the color distribution and color relationships in different areas of an image. In dynamic visual recognition, color features can be used to identify the colors of different objects, identify color changes in motion, etc.

[0058] Shape features: Shape is one of the important features of an object. Shape features describe the geometric shape of the object's outline or boundary. In dynamic visual recognition, shape features can be used to distinguish different object shapes, such as identifying the shape changes of objects in motion;

[0059] In order to maintain the effectiveness of the features, the dimensionality of the image features needs to be reduced. In the feature extraction process, dimensionality reduction techniques such as principal component analysis (PCA) or t-distributed neighbor embedding (t-SNE) are usually used to ensure that the number of features is reduced while retaining key information.

[0060] To extract features for image processing and pattern recognition, each pixel in the image is compared with its neighboring pixels using the local binary pattern (LBP), and a binary pattern is generated from the comparison results. Then, the entire image is scanned to extract the local binary pattern of each pixel, and finally a feature vector is formed. The specific process is as follows:

[0061] Select a neighborhood. For each pixel in the image, select a neighborhood around it, such as an 8-neighborhood or a 16-neighborhood.

[0062] Binary comparison: compare the gray value of the central pixel with the gray value of each pixel in the neighborhood to obtain a binary sequence. If the gray value of the neighborhood pixel is greater than or equal to the gray value of the central pixel, the binary bit at the corresponding position is set to s=1, otherwise it is set to s=0;

[0063] Generate a pattern, convert the obtained binary sequence into a binary number as the local binary pattern of the pixel;

[0064] Feature extraction, scan the entire image, extract the local binary pattern of each pixel, and form a feature vector; Where P is the total number of pixels in the region, g c is the gray value of the center pixel, g pIt is the local pixel value of pixel point p. The corresponding local binary pattern is calculated for each pixel in the image to form a new image or feature vector for subsequent image analysis and pattern recognition.

[0065] When using the LBP operator to extract features, the statistical histogram of the LBP pattern is used to represent the image information. When too many pattern types are used, the amount of data is huge, making the histogram too sparse, affecting the representation effect of the features. Therefore, the original LBP pattern is reduced in dimension to reduce the amount of data while maintaining the effective expression of image information. The steps of using principal component analysis (PCA) to reduce the dimension of the LBP pattern are as follows:

[0066] Collect image samples and use the LBP operator to extract the LBP features of each image. The LBP features of all images are combined into a matrix, where each row represents the LBP feature vector of an image;

[0067] The LBP feature matrix is ​​mean normalized, that is, the value of each column is subtracted from the mean of the corresponding column to ensure that the center of the data is at the origin;

[0068] Calculate the covariance matrix of the LBP feature matrix. The covariance matrix describes the correlation between different features.

[0069] Perform eigenvalue decomposition on the covariance matrix to obtain the corresponding eigenvalues ​​and corresponding eigenvectors, and use the eigenvectors to form a new eigenspace, where each eigenvector corresponds to a principal component;

[0070] The eigenvectors are sorted according to the size of the eigenvalues, and the first k eigenvectors are selected as principal components, where k is the dimension after dimensionality reduction. The eigenvector corresponding to the principal component contains the largest variance in the data. The original LBP feature matrix is ​​projected into the low-dimensional space composed of the selected principal components, that is, it is obtained by multiplying the LBP feature matrix with the selected principal component matrix, thereby obtaining the characteristic data of dynamic streaming media.

[0071] For example, assuming that the original LBP feature matrix is ​​X (assuming that each row represents an image and each column represents an LBP feature), the mean normalized matrix is ​​Z, the covariance matrix is ​​C, and the number of principal components selected is k, then the feature matrix after dimensionality reduction is Y;

[0072] The mean normalization formula is: Z = X-mean(X), where mean(X) means calculating the mean for each column in the matrix X. If X is a matrix containing multiple samples, each row represents a sample, and each column represents a feature (LBP feature), then mean(X) will calculate the mean of each column to obtain a row vector in which each element is the mean of the corresponding column. The covariance matrix formula is: C = 1 / (N-1)*Z*Z T , where ZT represents the transpose of the normalized matrix Z. The calculation formula for selecting the principal component is: Y = Z * V k , where V k The matrix Y contains the eigenvectors corresponding to the first k eigenvalues, which is the LBP feature matrix after dimensionality reduction.

[0073] For the defined computer-generated art, including the generated content, style, form, etc., for example, it can be defined as generating abstract art paintings. The collected and processed data are pre-processed, including image resizing, normalization, noise processing, etc., to ensure the quality and consistency of the data set again;

[0074] Use generative adversarial networks for artistic image generation tasks. Deep Convolutional GAN ​​(DCGAN) is a variant of generative adversarial networks for image generation tasks. The DCGAN model includes the following components:

[0075] Generator component structure: Input a random noise vector (usually uniformly distributed or normally distributed), use the deconvolution layer (Transpose Convolution) for upsampling, gradually map the noise vector to an image similar to the real image, and usually use activation functions such as LeakyReLU to generate the initial artistic image;

[0076] The structural logic of the generator is as follows: for each layer in the generator, it can be represented as a random noise input layer: x0=z, where z represents the input random noise vector, the initial deconvolution layer: x1=LeakyReLU(BatchNorm(W1*x0+b0)), the rest of the deconvolution layers are similar, and the final output layer outputs the art image: IMG=Tanh(BatchNorm(W n *x n-1 +b n-1 )),W n , b n-1 Represents the weight matrix and bias vector of the nth convolutional layer, LeakyReLU is a rectified linear unit with leakage, BatchNorm is batch normalization, and Tanh is a hyperbolic tangent activation function

[0077] Discriminator component structure: a generated art image or a real art image, the convolutional layer performs downsampling, gradually maps the image to a binary classification result, that is, whether it is a real image or a generated image, and then the batch normalization layer is used to improve the training stability and output the binary classification result (true or false);

[0078] The discriminator and the generator have the same structural logic. Through adversarial training, the generator and the discriminator compete with each other, so that the generator gradually generates more realistic artistic images, while the discriminator strives to improve its ability to distinguish between real and generated images.

[0079] The optimization goal of the generator and discriminator is to minimize or maximize a specific loss function, such as binary cross entropy loss. The goal of the generator is to deceive the discriminator so that it cannot distinguish between generated images and real images. The goal of the discriminator is to accurately distinguish between generated images and real images. In addition, for different generative art tasks, it may also be necessary to consider the use of other structures and techniques, such as Conditional GAN ​​for conditional generation, or Style GAN for better control of the style of generated images. That is, when selecting the model structure, it is necessary to adjust and optimize it according to the characteristics of the specific task and data.

[0080] Analyze various data generated in the process of computer-generated art images to determine the accuracy of visual recognition, that is, obtain style determination information and tuning influence information in the process of computer-generated art images;

[0081] The style determination information includes the structural quality similarity index and is calibrated as JGZ, and the tuning impact information includes the expression stability index and is calibrated as BDL;

[0082] The structural quality similarity index in the style determination information is used to measure the similarity between the generated image and the expected image. In the process of computer-generated art, the structural quality similarity index is mainly used to describe the structural similarity between the generated art image and the expected image or artistic style. By analyzing the structural similarity, the structural quality of the generated art image can be evaluated to ensure that the generated image matches the artistic style or expected image in structure, thereby maintaining the consistency of the artistic style, retaining image details and ensuring image clarity. It helps to quantify and analyze the structural characteristics of the generated art, thereby optimizing the generation process. The structural quality similarity index will have an effect on the following aspects:

[0083] Contrast consistency: The structural quality similarity index takes into account the contrast similarity of the images, ensuring that the generated art image maintains a consistent contrast level with the expected image. The structural quality of the generated image can be improved based on the structural quality similarity index, maintaining more clarity and details, making the generated art more realistic.

[0084] Structural similarity: The structural quality similarity index not only considers the overall brightness and contrast, but also includes an assessment of the similarity of image structure, ensuring that the generated art structurally matches the artistic style or intended image, which helps maintain the consistency of the overall artwork and ensures that the generated art has better visual effects.

[0085] The structural quality similarity index is obtained as follows:

[0086] Get a representative sample image from the sample data to establish a sample image set TXi={TX1, TX2, ..., TXn}, where n is a positive integer, get the expected generated image YQ, and calculate the pixel average value of the sample image set: Pixel variance of a sample image collection: Similarly, calculate the pixel average Yavg and pixel variance of the expected generated image Calculate the difference: The structural quality similarity index is calculated and the calculation expression is:

[0087] It should be noted that obtaining representative sample images in the sample data refers to representative samples that cover the diversity and complexity required for generating computer art tasks, which are representative among multiple similar samples and have a large weight influence on the generated results, thereby reducing the increase in load caused by analyzing too many sample images; calculating the average value of the sample image helps to understand the overall distribution characteristics of the original data, so as to serve as a reference for style transfer, grasp the brightness center trend of the original sample, and help the generated art image to maintain a similar brightness distribution as the original sample on the whole, and is used to adjust the contrast and structural features of the generated image to ensure that the generated art is more consistent with the original sample in structure. The average value of the sample image can be used as a reference point when generating art images, which helps to ensure that the generated art image is similar to the original sample on the whole, maintains consistent brightness and style characteristics, so as to improve the quality and realism of the generated image.

[0088] The expression stability index in the tuning impact information indicates the quality stability of the results generated by the generated artwork in different situations, thereby helping to evaluate the computing resource consumption, actual generation quality, and expressiveness of computer-generated art. Further optimization and adjustment are performed based on the analysis results to ensure that in actual applications, computing resources can be used efficiently and artworks with good quality and expressiveness can be generated. The expression stability index affects the following aspects:

[0089] The expression stability index is directly related to the expressive quality of generated artworks in different situations. A high expression stability index means that the generated artworks can maintain good expressiveness under different conditions, while a low index may indicate poor expressiveness in certain situations. Stability evaluation of the expressiveness of artworks helps ensure that the generative model can consistently provide high-quality artistic expression in various artistic styles, forms, and themes.

[0090] By adjusting model parameters, architecture or training strategies to improve expressive stability, the performance of the model can be optimized and computing resources can be used more efficiently. Efficient use of computing resources is a key factor in ensuring feasibility in practical applications. The expression stability index can provide guidance in balancing quality and resource consumption, thereby achieving the optimal control mode.

[0091] The expression stability index is obtained as follows:

[0092] Get the original image ST and the generated image SC, calculate the error value of each original image, and the error value calculation expression is: m is a positive integer, which indicates the number of pixels in the generated image. W and H are the width and height of the image respectively. The error value set WCv = {WC1, WC2, ..., WCr} of the original image is established. r is a positive integer, and the error mean is obtained. Get the maximum possible signal-to-noise ratio MAX of the pixel and calculate the signal-to-noise ratio of the pixel value: Get the expected time SJ and actual time SJ of generating images 实 , calculate the expression stability index, the calculation expression is:

[0093] It should be noted that, according to the number of image samples, an appropriate number of samples of different sizes are selected as original images and analyzed.

[0094] The obtained structural quality similarity index JGZ and expression stability index BDL are comprehensively calculated to obtain the identification generation accuracy coefficient, which is expressed as: In the formula, S x To identify and generate accurate coefficients, α and β are preset proportional coefficients of the structural quality similarity index JGZ and the expression stability index BDL, and both α and β are greater than 0.

[0095] It should be noted that the size of the preset proportional coefficient is a specific value obtained by quantifying each parameter. In order to facilitate subsequent comparison, the size of the coefficient depends on the amount of sample data and the preliminary setting of the corresponding preset proportional coefficient for each group of sample data by technical personnel in this field. It is not unique, as long as it does not affect the proportional relationship between the parameter and the quantized value, such as the expression stability index is proportional to the recognition generation precision coefficient.

[0096] The larger the similarity index of structural quality and the larger the expression stability index, that is, the larger the performance value of the recognition generation precision coefficient, indicates that when computer-generated art is generated through computer vision recognition, the generation efficiency and accuracy are higher, and the generated artworks are more consistent and stable in expression in multiple samples or multiple generations. The generation model can produce similar artistic effects under different conditions (different original images or inputs), the model has better adaptability to different inputs, and the generated artworks are of higher quality and reliability.

[0097] The smaller the structural quality similarity index and the expression stability index, that is, the smaller the performance value of the recognition generation accuracy coefficient, it means that the performance under different conditions is more likely to change greatly. As the input changes, the generated artwork is less similar to the original image, and the expression is not consistent under different conditions, which makes it more likely that the generated artwork will not be as good as expected.

[0098] Compare the generated recognition generation accuracy coefficient with the recognition generation evaluation threshold, generate different signals, and make corresponding adjustments based on the generated signals;

[0099] After obtaining the recognition generation accuracy coefficient, the recognition generation accuracy coefficient is compared with the recognition generation evaluation threshold;

[0100] If the recognition generation accuracy coefficient is greater than or equal to the recognition generation evaluation threshold, a recognition generation normal signal is generated, indicating that the generated artwork meets expectations, combines normally with the sample, is consistent in expression, and reaches the expected quality level;

[0101] If the recognition generation precision coefficient is less than the recognition generation evaluation threshold, a recognition generation fluctuation signal is generated, indicating that the generated artwork has fluctuating or unstable characteristics in quality and expression. The generation model has performance fluctuations, and attention should be paid to the stability and consistency of the model.

[0102] When the generation recognition generates a normal signal, follow the steps below to manage it:

[0103] Continuously monitor the performance of the generative model to ensure that the generated artwork maintains a normal level in quality and expression, record the generation recognition accuracy coefficient, evaluation threshold and other relevant information for long-term performance tracking, regularly evaluate the stability of the generative model, and analyze whether there are potential fluctuation factors, such as data changes, input changes, etc., and then consider adjusting model parameters, adding training data or improving the generation algorithm to improve the stability of the model. Based on the relevant performance information monitored, continuously optimize and improve the model to adapt to the generation needs under different conditions. Then, use the feedback mechanism to make adjustments based on user feedback and model output to continuously improve the effectiveness of the generative model;

[0104] At the same time, a notification mechanism needs to be established to promptly notify the system operation and maintenance team once an anomaly or performance degradation is detected so that they can respond quickly and troubleshoot the problem. The evaluation thresholds can then be updated to maintain the accuracy of the evaluation based on the evolution of the model and data.

[0105] It should be noted that the threshold settings in this embodiment are set according to actual conditions and are not fixed values, so no further analysis will be performed here.

[0106] Embodiment 2, this embodiment is a system embodiment of embodiment 1, which is used to implement the computer vision recognition method introduced in embodiment 1, such as Figure 2 As shown, specifically including:

[0107] The sample preprocessing module acquires sample data, obtains image information in different dimensions through visual recognition, extracts features from the acquired image information, compares pixels in the image with neighboring pixels using local binary patterns, determines the generated binary pattern, and reduces the dimension;

[0108] The recognition and analysis module uses a generative adversarial network to generate image tasks based on feature vectors, optimizes the generated images based on the generator and discriminator, determines the process of computer-generated art images and performs detection;

[0109] The management and control module analyzes various data generated in the process of computer-generated artistic images, determines the accuracy of visual recognition, and manages visual recognition based on the results.

[0110] The above formulas are all dimensionless and calculated numerically. Specific dimension removal can be achieved by various means such as standardization, which will not be elaborated here. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0111] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, an ATA hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state ATA hard disk.

[0112] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0113] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0114] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0115] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0116] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0117] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A computer vision recognition method, characterized in that: The steps include: Acquire sample data, obtain image information in different dimensions through visual recognition, and perform feature extraction on the acquired image information; Image processing and pattern recognition are performed during the feature extraction process, and the dimension of the image features is reduced. The pixels in the image are compared with the neighboring pixels using local binary patterns to determine the generated binary pattern; Extract local binary patterns of pixels according to binary patterns to form feature vectors, use generative adversarial networks to generate image tasks according to feature vectors, optimize generated images according to generators and discriminators, determine the process of computer-generated art images and perform detection; Analyze various data generated in the process of computer-generated artistic images, determine the accuracy of visual recognition, and manage visual recognition based on the results.

2. The computer vision recognition method according to claim 1, characterized in that: Use local binary patterns to compare pixels in the image with neighboring pixels to determine the resulting binary pattern. The specific process is as follows: Scan the entire image, and for each pixel in the image, select the neighborhood around it; Compare the grayscale value of the central pixel with the grayscale value of each pixel in the neighborhood to obtain a binary sequence; The obtained binary sequence is converted into a binary number as the local binary pattern of the pixel.

3. The computer vision recognition method according to claim 2, characterized in that: According to the binary pattern, the local binary pattern of the pixel point is extracted to form a feature vector. The specific process is as follows: Use the statistical histogram of LBP mode to represent image information; Collect image samples and use the LBP operator to extract the LBP features of each image. The LBP features of all images are combined into a matrix, where each row represents the LBP feature vector of an image. Normalize the LBP feature matrix by the mean, and subtract the mean of the corresponding column from the value of each column; Calculate the covariance matrix of the LBP feature matrix, and perform eigenvalue decomposition on the covariance matrix to obtain the corresponding eigenvalues ​​and corresponding eigenvectors, and use the eigenvectors to form a new feature space, where each eigenvector corresponds to a principal component; The original LBP feature matrix is ​​projected into the low-dimensional space composed of the selected principal components for dimensionality reduction.

4. The computer vision recognition method according to claim 3, characterized in that: Use the generative adversarial network to generate image tasks based on feature vectors, and optimize the generated images based on the generator and discriminator. The specific process is as follows: A random noise vector is input through the generator of the generative adversarial network, upsampled using the deconvolution layer, and the noise vector is mapped to a result similar to the real result; The generated result is downsampled by the convolution layer through the discriminator of the generative adversarial network, and the generated result is mapped to a binary classification result and output.

5. The computer vision recognition method according to claim 4, characterized in that: Analyze the data generated in the process of computer-generated art images to determine the accuracy of visual recognition. The specific process is as follows: Analyze various data generated in the process of computer-generated art images to obtain style determination information and tuning influence information in the process of computer-generated art images; The style determination information includes the structural quality similarity index, and the tuning impact information includes the expression stability index; The obtained structural quality similarity index and expression stability index are calculated to obtain the recognition generation accuracy coefficient; The structural quality similarity index and expression stability index are proportional to the recognition generation accuracy coefficient; The resulting recognition generation accuracy coefficient is compared to the recognition generation evaluation threshold.

6. The computer vision recognition method according to claim 5, characterized in that: The generated recognition generation accuracy coefficient is compared with the recognition generation evaluation threshold. The specific process is as follows: Compare the recognition generation accuracy coefficient with the recognition generation evaluation threshold; If the recognition generation accuracy coefficient is greater than or equal to the recognition generation evaluation threshold, a recognition generation normal signal is generated to maintain normal management; If the recognition generation accuracy coefficient is less than the recognition generation evaluation threshold, an recognition generation fluctuation signal is generated and regulated and managed.

Images