Image processing method and system based on computer vision

Abstract

The invention is suitable for the field of image processing, and provides an image processing method and system based on computer vision, and the method comprises the steps: collecting a historical satisfactory revised image of a targeted user, and extracting a revised image parameter corresponding to the historical satisfactory revised image; obtaining a new selfie image, performing face detection on the new selfie image, and obtaining face key feature point location information; matching the face key feature point location information of the new selfie image with the retouching parameters corresponding to the historical satisfactory retouching image, and generating a processing matching screening result; on the basis of the processing matching screening result, additional beautification processing is conducted on the new selfie image, and an additional beautified image is generated. According to the method and the system, the quality of beautification processing is remarkably improved by utilizing the historical image retouching data and introducing the generative adversarial network. Compared with a traditional facial beautification algorithm, the method can more accurately match user demands, and avoids excessive or insufficient facial beautification processing. And optimizing graph retouching parameters through the generative adversarial network.

Description

Technical Field

[0001] The present invention belongs to the field of image processing, and particularly relates to an image processing method and system based on computer vision. Background Art

[0002] With the popularization of smart phones, selfies have become an important way for people to record their lives and share their daily lives. Users' demand for the beauty effect of selfie images is increasing day by day, and they expect to obtain photos with professional photo retouching standards through simple operations.

[0003] In recent years, with the rapid development of computer vision technology, new ideas and methods have been brought to the beauty processing of selfie images. The beauty algorithm based on deep learning can analyze and process different facial features and image contents more accurately through the learning of a large amount of data. However, the existing deep learning-based beauty methods often rely on large-scale publicly available datasets for training, and there are certain differences between these datasets and users' actual selfie scenarios and personal aesthetics. Moreover, the training process usually requires a large amount of computing resources and time costs, and it is difficult to respond quickly and adjust flexibly in practical applications.

[0004] In order to overcome the deficiencies of the prior art, the present invention proposes an image processing method based on computer vision to realize selfie image processing that better meets users' needs and aesthetic preferences. Summary of the Invention

[0005] The purpose of the present invention is to provide an image processing method and system based on computer vision, aiming to solve the problems raised in the above background art.

[0006] The present invention is implemented as follows. On the one hand, an image processing method based on computer vision, the method includes:

[0007] Collect the historical satisfactory retouched images of the targeted user, and extract the retouching parameters corresponding to the historical satisfactory retouched images;

[0008] Obtain a new selfie image, perform face detection on the new selfie image, and obtain the key feature point position information of the face;

[0009] Match the key feature point position information of the face in the new selfie image with the retouching parameters corresponding to the historical satisfactory retouched images to generate a processing matching and screening result;

[0010] Based on the processing matching and screening result, perform additional beauty processing on the new selfie image to generate an additional beauty image;

[0011] During the process of performing additional beauty processing on the new selfie image, the generative adversarial network optimizes the retouching parameters corresponding to the historical satisfactory retouched images.

[0012] As a further solution of the present invention, the acquisition of the new self-taken image, face detection of the new self-taken image, and acquisition of the key facial feature point information specifically include:

[0013] Perform multi-scale downsampling processing on the new self-taken image to obtain image features at different resolutions;

[0014] Use the sliding window algorithm to screen the face probability region based on the image features at different resolutions;

[0015] For the face probability region, locate the positions of the key facial organs;

[0016] Obtain the positions of the key facial organs and generate the facial contour;

[0017] Encode the positions of the key facial organs and the facial contour features to generate a target feature vector.

[0018] As a further solution of the present invention, the matching of the key facial feature point information of the new self-taken image with the corresponding retouching parameters of the historical satisfactory retouched image to generate a processing matching and screening result specifically includes:

[0019] Perform dimensionality reduction processing on the target feature vector;

[0020] Use the cosine similarity algorithm to calculate the similarity between the historical satisfactory retouched image and the feature vector after dimensionality reduction of the target feature vector;

[0021] Combine the Euclidean distance algorithm to perform secondary verification on the cosine similarity calculation result;

[0022] According to the similarity and distance calculation results, sort the corresponding retouching parameters of the historical satisfactory retouched image in descending order, and screen out several corresponding retouching parameters whose similarity with the features of the new self-taken image is greater than the similarity threshold.

[0023] As a further solution of the present invention, the additional beauty processing of the new self-taken image based on the processing matching and screening result to generate an additional beauty image specifically includes:

[0024] Based on the several selected corresponding retouching parameters, determine the initial parameter settings of the new image beauty algorithm model;

[0025] Based on the image fusion algorithm with an attention mechanism, fuse the determined initial parameter settings into the new self-taken image;

[0026] Real-time monitor the color balance and contrast changes of the new self-taken image;

[0027] Collect the adjustment instructions input by the target user in real time and dynamically interact to adjust the parameter settings of the beauty algorithm model.

[0028] As a further solution of the present invention, during the process of performing additional beauty processing on the new self-taken image, introducing a generative adversarial network to optimize the corresponding image retouching parameters of the historical satisfactory retouched images specifically includes:

[0029] Construct the network structures of the generator and discriminator, initialize the weight parameters and set the hyperparameters;

[0030] Extract samples from several corresponding image retouching parameters selected to form a training set and perform normalization preprocessing;

[0031] During iterative training, alternately fix the discriminator and the generator, and input the new self-taken image to update the parameters of both;

[0032] Based on the changes in the losses of the generator and discriminator, dynamically adjust the hyperparameters and evaluate the quality of the new self-taken image;

[0033] After the training is completed, import several corresponding image retouching parameters selected into the generator to obtain the beauty optimization result and use it for the beauty processing of the new self-taken image.

[0034] As a further solution of the present invention, on the other hand, an image processing system based on computer vision, the system includes:

[0035] An acquisition module, used to acquire the historical satisfactory retouched images of the targeted user,

[0036] An extraction module, used to extract the corresponding image retouching parameters of the historical satisfactory retouched images;

[0037] A first acquisition module, used to acquire the new self-taken image;

[0038] A face detection module, used to perform face detection on the new self-taken image;

[0039] A second acquisition module, used to acquire the key feature point information of the face;

[0040] A matching module, used to match the key feature point information of the face in the new self-taken image with the corresponding image retouching parameters of the historical satisfactory retouched images;

[0041] A first generation module, used to generate the processing matching and screening results;

[0042] A beauty processing module, used to perform additional beauty processing on the new self-taken image based on the processing matching and screening results;

[0043] A second generation module, used to generate the additional beauty image;

[0044] A generative adversarial network module, used to optimize the corresponding image retouching parameters of the historical satisfactory retouched images by the generative adversarial network during the process of performing additional beauty processing on the new self-taken image.

[0045] As a further solution of the present invention, the matching module specifically includes:

[0046] A dimensionality reduction processing unit for performing dimensionality reduction processing on the target feature vector;

[0047] A calculation unit for calculating the similarity between the historical satisfactory retouched image and the feature vector after dimensionality reduction of the target feature vector using the cosine similarity algorithm;

[0048] A verification unit for performing secondary verification on the cosine similarity calculation result in combination with the Euclidean distance algorithm;

[0049] A sorting unit for sorting the retouching parameters corresponding to the historical satisfactory retouched images in descending order according to the similarity and distance calculation results;

[0050] A screening unit for screening out several corresponding retouching parameters whose feature similarity with the new self-taken image is greater than the similarity threshold.

[0051] As a further solution of the present invention, the beauty processing module specifically includes:

[0052] A determination unit for determining the initial parameter settings of the new image beauty algorithm model based on the several corresponding retouching parameters screened out;

[0053] A fusion unit for fusing the determined initial parameter settings into the new self-taken image based on the image fusion algorithm based on the attention mechanism;

[0054] A real-time monitoring unit for real-time monitoring of the color balance and contrast changes of the new self-taken image;

[0055] An acquisition unit for acquiring the adjustment instructions input by the target user in real time;

[0056] A dynamic interaction unit for dynamically interacting to adjust the parameter settings of the beauty algorithm model.

[0057] An image processing method and system based on computer vision provided by the present invention. By using historical retouching data and introducing an adversarial generation network, the quality of beauty processing is significantly improved. Compared with traditional beauty algorithms, it can more accurately match user needs and avoid excessive or insufficient beauty processing. By optimizing the retouching parameters through the adversarial generation network, the beauty effect is made more natural and real, meeting the user's pursuit of personalized and high-quality beauty. Description of the Drawings

[0058] Figure 1 It is the main flowchart of an image processing method based on computer vision.

[0059] Figure 2It is a flowchart for obtaining a new self - taken image in an image - processing method based on computer vision, performing face detection on the new self - taken image, and obtaining information on key facial feature points.

[0060] Figure 3 It is a flowchart for matching the information on key facial feature points of the new self - taken image with the corresponding retouching parameters of the historical satisfactory retouched image in an image - processing method based on computer vision to generate a processed matching and screening result.

[0061] Figure 4 It is a flowchart for performing additional beauty processing on the new self - taken image based on the processed matching and screening result to generate an additional beauty image in an image - processing method based on computer vision.

[0062] Figure 5 It is a flowchart for introducing an adversarial generative network to optimize the corresponding retouching parameters of the historical satisfactory retouched image during the process of performing additional beauty processing on the new self - taken image in an image - processing method based on computer vision.

[0063] Figure 6 It is the main structure diagram of an image - processing system based on computer vision.

[0064] Figure 7 It is the structural block diagram of the matching module in an image - processing system based on computer vision.

[0065] Figure 8 It is the structural block diagram of the beauty - processing module in an image - processing system based on computer vision. Detailed implementation manners

[0066] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0067] The following describes the specific implementation of the present invention in detail with reference to specific embodiments.

[0068] The image - processing method and system based on computer vision provided by the present invention solve the technical problems in the background art.

[0069] As Figure 1 shown, it is the main flowchart of an image - processing method based on computer vision provided by an embodiment of the present invention. The image - processing method based on computer vision includes:

[0070] Step S100: Collect the historical satisfactory retouched images of the targeted user and extract the corresponding retouching parameters of the historical satisfactory retouched images;

[0071] Step S200: Obtain a new self - portrait image, perform face detection on the new self - portrait image, and obtain the information of key facial feature points;

[0072] Step S300: Match the information of key facial feature points of the new self - portrait image with the corresponding retouching parameters of the historical satisfactory retouched image, and generate a processing matching and screening result;

[0073] Step S400: Based on the processing matching and screening result, perform additional beauty processing on the new self - portrait image to generate an additional beauty image;

[0074] Step S500: During the process of performing additional beauty processing on the new self - portrait image, introduce an adversarial generative network to optimize the corresponding retouching parameters of the historical satisfactory retouched image;

[0075] When this embodiment is applied, first, accurately collect the historical satisfactory retouched images of the targeted users, and at the same time deeply extract the corresponding retouching parameters of these images, covering key information such as skin - smoothing intensity and whitening degree, to build a personalized historical retouching database. Then, when a new self - portrait image is obtained, use advanced computer vision technology to perform face detection on it and accurately obtain the information of key facial feature points. Subsequently, match the information of key facial feature points of the new self - portrait image with the retouching parameters in the historical retouching database, and through algorithm analysis, generate a processing matching and screening result. When performing additional beauty processing on the new self - portrait image, introduce an adversarial generative network. Utilize the adversarial training mechanism of the generator and discriminator to optimize the retouching parameters corresponding to the historical satisfactory retouched image, so that the image after beauty processing has both a sense of reality and natural beauty, and finally generate a high - quality additional beauty image.

[0076] As Figure 2 shown, as a preferred embodiment of the present invention, the obtaining of the new self - portrait image, performing face detection on the new self - portrait image, and obtaining the information of key facial feature points specifically include:

[0077] Step S201: Perform multi - scale downsampling processing on the new self - portrait image to obtain image features at different resolutions;

[0078] Step S202: Use the sliding window algorithm to screen the face probability region based on the image features at different resolutions;

[0079] Step S203: In the face probability region, locate the positions of key facial organs;

[0080] Step S204: Obtain the positions of key facial organs and generate a facial contour;

[0081] Step S205: Encode the positions of key facial organs and facial contour features to generate a targeted feature vector;

[0082] When this embodiment is applied, first, multi-scale downsampling processing is performed on the new self-taken image. Through this operation, image features at different resolutions can be obtained, and this multi-resolution information provides rich image details and levels for subsequent processing. Then, the sliding window algorithm is used to screen out the face probability regions based on the image features at different resolutions obtained previously, improving the accuracy and robustness of face detection. Subsequently, the positions of key facial organs are located in the screened face probability regions, and the position information of important organs such as eyes, nose, and mouth is accurately found. Further, the facial contour is generated based on these key organ positions, improving the overall structural information of the face. Finally, the positions of key facial organs and the facial contour features are encoded to generate a target feature vector, converting the complex facial information into a simple and easy-to-process data form.

[0083] As Figure 3 shown, as a preferred embodiment of the present invention, the matching of the key feature point information of the new self-taken image face with the corresponding retouching parameters of the historical satisfactory retouched image to generate a processing matching and screening result specifically includes:

[0084] Step S301: Perform dimensionality reduction processing on the target feature vector;

[0085] Step S302: Use the cosine similarity algorithm to calculate the similarity between the historical satisfactory retouched image and the feature vector after dimensionality reduction of the target feature vector;

[0086] Step S303: Combine the Euclidean distance algorithm to perform secondary verification on the cosine similarity calculation result;

[0087] Step S304: According to the similarity and distance calculation results, sort the corresponding retouching parameters of the historical satisfactory retouched image in descending order, and screen out several corresponding retouching parameters whose similarity with the new self-taken image features is greater than the similarity threshold;

[0088] When this embodiment is applied, first, the target feature vector of the generated new self-taken image is dimensionally reduced to reduce the data dimension, retain key information while reducing the subsequent calculation amount. Subsequently, the cosine similarity algorithm is used to calculate the similarity between the historical satisfactory retouched image and the feature vector after the dimensional reduction of the target feature vector to evaluate the similarity between the two. To improve the accuracy of the matching, the Euclidean distance algorithm is also combined to perform a secondary verification on the cosine similarity calculation result. Through the synergistic effect of the two algorithms, the association between the two is measured more accurately. Finally, according to the similarity and distance calculation results, the retouching parameters corresponding to the historical satisfactory retouched images are sorted in descending order, and several corresponding retouching parameters with similarity greater than the set threshold are selected. This screening process can ensure that the historical retouching parameters most similar to the features of the new self-taken image are selected, providing a reliable data basis for subsequent image processing operations based on historical experience, effectively utilizing the user's historical satisfactory retouching data, making the image processing process more personalized, improving the quality and stability of the processing effect, and providing important data support for operations such as image adjustment that meet the user's personalized needs.

[0089] As Figure 4 shown, as a preferred embodiment of the present invention, the additional beauty processing of the new self-taken image based on the processing matching and screening results to generate an additional beauty image specifically includes:

[0090] Step S401: Based on the selected several corresponding retouching parameters, determine the initial parameter settings of the new image beauty algorithm model;

[0091] Step S402: Based on the image fusion algorithm with an attention mechanism, fuse the determined initial parameter settings into the new self-taken image;

[0092] Step S403: Real-time monitor the color balance and contrast changes of the new self-taken image;

[0093] Step S404: Collect the adjustment instructions input by the target user in real time and dynamically interact to adjust the parameter settings of the beauty algorithm model.

[0094] It should be understood that, first, based on a number of corresponding retouching parameters selected previously, initial parameter settings are determined for the new image beauty algorithm model to ensure that subsequent processing can make full use of historical data and conform to user preferences. Then, using an image fusion algorithm based on the attention mechanism, the initial parameter settings are skillfully fused into the new self-taken image. This algorithm helps to accurately apply parameter information to the corresponding parts of the image to achieve targeted beauty effects. During the processing, the color balance and contrast changes of the new self-taken image are monitored in real time to ensure that the overall visual effect of the image is in the optimal state. In addition, adjustment instructions input by the targeted user in real time can be collected, and the parameter settings of the beauty algorithm model are adjusted through dynamic interaction, enabling the user to flexibly adjust the beauty effect according to their immediate needs, thus achieving highly personalized and dynamic self-taken image beauty processing.

[0095] As Figure 5 shown, as a preferred embodiment of the present invention, in the process of performing additional beauty processing on the new self-taken image, the optimization of the corresponding retouching parameters for the historical satisfactory retouched image by introducing an adversarial generative network specifically includes:

[0096] Step S501: Construct the network structures of the generator and discriminator, initialize the weight parameters and set the hyperparameters;

[0097] Step S502: Extract samples from a number of corresponding retouching parameters selected to form a training set and perform normalization preprocessing;

[0098] Step S503: In iterative training, alternately fix the discriminator and the generator, and input the new self-taken image to update the parameters of both;

[0099] Step S504: Dynamically adjust the hyperparameters based on the loss changes of the generator and discriminator and evaluate the quality of the new self-taken image;

[0100] Step S505: After the training is completed, import a number of corresponding retouching parameters selected into the generator to obtain the beauty optimization result and use it for the beauty of the new self-taken image;

[0101] When this embodiment is applied, first, the network structures of the generator and the discriminator are constructed. At the same time, the weight parameters are initialized and the hyperparameters are reasonably set to lay a foundation for subsequent training. Then, samples are extracted from the selected multiple corresponding retouching parameters to construct a training set, and it is preprocessed by normalization to ensure the consistency and effectiveness of the data. During the training process, an iterative training method is adopted, where the discriminator and the generator are alternately fixed, and the new self-taken image is used as the input to update the parameters of both. Through the adversarial training of both, the network performance is continuously optimized. During this period, the hyperparameters are dynamically adjusted according to the loss changes of the generator and the discriminator, and the quality of the new self-taken image is evaluated as the basis for adjustment to ensure that the training develops in the direction of improving the image quality. Finally, after the training is completed, the selected retouching parameters are imported into the generator to obtain the beauty optimization result, and the beauty optimization result will be applied to the beauty processing of the new self-taken image. Through the unique architecture and dynamic training process of GAN, this technology can effectively utilize historical retouching parameters and continuously optimize the beauty effect of new self-taken images.

[0102] As Figure 6 shown, as another preferred embodiment of the present invention, on the other hand, an image processing system based on computer vision, the system includes:

[0103] An acquisition module 100 for acquiring the historical satisfactory retouched images of the targeted user,

[0104] An extraction module 200 for extracting the retouching parameters corresponding to the historical satisfactory retouched images;

[0105] A first acquisition module 300 for acquiring a new self-taken image;

[0106] A face detection module 400 for performing face detection on the new self-taken image;

[0107] A second acquisition module 500 for acquiring the key feature point information of the face;

[0108] A matching module 600 for matching the key feature point information of the face in the new self-taken image with the retouching parameters corresponding to the historical satisfactory retouched images;

[0109] A first generation module 700 for generating a processing matching and screening result;

[0110] A beauty processing module 800 for performing additional beauty processing on the new self-taken image based on the processing matching and screening result;

[0111] A second generation module 900 for generating an additional beauty image;

[0112] An adversarial generation network module 1000 for optimizing the retouching parameters corresponding to the historical satisfactory retouched images by the adversarial generation network during the process of performing additional beauty processing on the new self-taken image.

[0113] When this embodiment is applied, the acquisition module 100 acquires the historical satisfactory retouched images of the targeted user, the extraction module 200 extracts the retouching parameters corresponding to the historical satisfactory retouched images, the first acquisition module 300 acquires the new self-taken images, the face detection module 400 performs face detection on the new self-taken images, the second acquisition module 500 acquires the key feature point information of the face, the matching module 600 matches the key feature point information of the face in the new self-taken images with the retouching parameters corresponding to the historical satisfactory retouched images, the first generation module 700 generates the processed matching and screening results, and based on the processed matching and screening results, the beauty processing module 800 performs additional beauty processing on the new self-taken images, and the second generation module 900 generates the additional beauty images. During the process of performing additional beauty processing on the new self-taken images, the adversarial generation network module 1000 optimizes the retouching parameters corresponding to the historical satisfactory retouched images through the adversarial generation network.

[0114] As Figure 7 shown, as another preferred embodiment of the present invention, the matching module 600 specifically includes:

[0115] The dimensionality reduction processing unit 601 is used to perform dimensionality reduction processing on the targeted feature vector;

[0116] The calculation unit 602 is used to calculate the similarity between the historical satisfactory retouched image and the feature vector after dimensionality reduction of the targeted feature vector using the cosine similarity algorithm;

[0117] The verification unit 603 is used to perform secondary verification on the cosine similarity calculation result by combining the Euclidean distance algorithm;

[0118] The sorting unit 604 is used to sort the retouching parameters corresponding to the historical satisfactory retouched images in descending order according to the similarity and distance calculation results;

[0119] The screening unit 605 is used to screen out several corresponding retouching parameters whose feature similarity with the new self-taken image is greater than the similarity threshold.

[0120] When this embodiment is applied, the dimensionality reduction processing unit 601 performs dimensionality reduction processing on the targeted feature vector, the calculation unit 602 calculates the similarity between the historical satisfactory retouched image and the feature vector after dimensionality reduction of the targeted feature vector using the cosine similarity algorithm, the verification unit 603 performs secondary verification on the cosine similarity calculation result by combining the Euclidean distance algorithm, the sorting unit 604 sorts the retouching parameters corresponding to the historical satisfactory retouched images in descending order according to the similarity and distance calculation results, and the screening unit 605 screens out several corresponding retouching parameters whose feature similarity with the new self-taken image is greater than the similarity threshold.

[0121] As Figure 8As shown, as another preferred embodiment of the present invention, the beauty processing module 800 specifically includes:

[0122] A determination unit 801, configured to determine an initial parameter setting of a new image beauty algorithm model based on a plurality of selected corresponding image retouching parameters;

[0123] A fusion unit 802, configured to fuse the determined initial parameter setting into the new self-taken image based on an image fusion algorithm based on an attention mechanism;

[0124] A real-time monitoring unit 803, configured to real-time monitor the color balance and contrast change of the new self-taken image;

[0125] An acquisition unit 804, configured to acquire an adjustment instruction input by the targeted user in real time;

[0126] A dynamic interaction unit 805, configured to dynamically interact to adjust the parameter setting of the beauty algorithm model.

[0127] When this embodiment is applied, based on a plurality of selected corresponding image retouching parameters, the determination unit 801 determines an initial parameter setting of a new image beauty algorithm model. Based on an image fusion algorithm based on an attention mechanism, the fusion unit 802 fuses the determined initial parameter setting into the new self-taken image. The real-time monitoring unit 803 real-time monitors the color balance and contrast change of the new self-taken image. The acquisition unit 804 acquires an adjustment instruction input by the targeted user in real time. The dynamic interaction unit 805 dynamically interacts to adjust the parameter setting of the beauty algorithm model.

[0128] In the above embodiments of the present invention, an image processing method based on computer vision is provided, and an image processing system based on computer vision is also provided. First, accurately collect the historical satisfactory retouched images of the target users, and at the same time deeply extract the retouching parameters corresponding to these images, covering key information such as skin smoothing intensity and whitening degree, to build a personalized historical retouching database. Then, when a new self-taken image is obtained, use advanced computer vision technology to perform face detection on it, and accurately obtain the key feature point information of the face. Subsequently, match the key feature point information of the face in the new self-taken image with the retouching parameters in the historical retouching database, and generate a processing matching and screening result through algorithm analysis. When performing additional beauty processing on the new self-taken image, introduce an adversarial generative network. Utilize the adversarial training mechanism of the generator and the discriminator to optimize the retouching parameters corresponding to the historical satisfactory retouched images, so that the images after beauty processing have both a sense of reality and natural beauty, and finally generate high-quality additional beauty images; this method and system significantly improve the quality of beauty processing by using historical retouching data and introducing an adversarial generative network. Compared with traditional beauty algorithms, it can more accurately match the user's needs and avoid excessive or insufficient beauty processing. By optimizing the retouching parameters through the adversarial generative network, the beauty effect is more natural and real, meeting the user's pursuit of personalized and high-quality beauty.

[0129] In order to enable the above method and system to run smoothly, in addition to including the above various modules, the system may also include more or fewer components than those described above, or combine certain components, or different components. For example, it may include input and output devices, network access devices, buses, processors, and memories, etc.

[0130] The so-called processor may be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor, etc. The above processor is the control center of the above system, connecting various parts through various interfaces and lines.

[0131] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0132] The above-described embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all fall within the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the appended claims.

[0133] The foregoing is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An image processing method based on computer vision, characterized in that, The method includes: Collecting the historical satisfactory retouched images of the targeted user and extracting the corresponding retouching parameters of the historical satisfactory retouched images; Obtaining a new self-taken image, performing face detection on the new self-taken image, and obtaining the information of key facial feature points; Matching the information of key facial feature points of the new self-taken image with the corresponding retouching parameters of the historical satisfactory retouched images to generate a processed matching and screening result; Based on the processed matching and screening result, performing additional beauty retouching on the new self-taken image to generate an additional beauty retouched image; During the process of performing additional beauty retouching on the new self-taken image, the adversarial generation network optimizes the corresponding retouching parameters of the historical satisfactory retouched images.

2. The image processing method based on computer vision according to claim 1, wherein, The obtaining of the new self-taken image, performing face detection on the new self-taken image, and obtaining the information of key facial feature points specifically includes: Performing multi-scale downsampling processing on the new self-taken image to obtain image features at different resolutions; Using the sliding window algorithm, based on the image features at different resolutions, screening the face probability regions; For the face probability regions, locating the positions of key facial organs; Obtaining the positions of key facial organs and generating a facial contour; Encoding the positions of key facial organs and the facial contour features to generate a targeted feature vector.

3. The image processing method based on computer vision according to claim 2, wherein The matching of the information of key facial feature points of the new self-taken image with the corresponding retouching parameters of the historical satisfactory retouched images to generate a processed matching and screening result specifically includes: Performing dimensionality reduction processing on the targeted feature vector; Using the cosine similarity algorithm to calculate the similarity between the historical satisfactory retouched image and the feature vector after dimensionality reduction of the targeted feature vector; Combining with the Euclidean distance algorithm to perform secondary verification on the calculation result of the cosine similarity; According to the similarity and distance calculation results, sorting the corresponding retouching parameters of the historical satisfactory retouched images in descending order, and screening out several corresponding retouching parameters whose similarity with the features of the new self-taken image is greater than the similarity threshold.

4. The image processing method based on computer vision according to claim 1, characterized in that The performing of additional beauty retouching on the new self-taken image based on the processed matching and screening result to generate an additional beauty retouched image specifically includes: Based on the several corresponding retouching parameters screened out, determining the initial parameter settings of the new image beauty algorithm model; Based on the image fusion algorithm with an attention mechanism, fusing the determined initial parameter settings into the new self-taken image; Real-time monitoring the color balance and contrast changes of the new self-taken image; Collecting the adjustment instructions input by the targeted user in real time and dynamically interacting to adjust the parameter settings of the beauty algorithm model.

5. The image processing method based on computer vision according to claim 3, characterized in that, The introducing of the adversarial generation network to optimize the corresponding retouching parameters of the historical satisfactory retouched images during the process of performing additional beauty retouching on the new self-taken image specifically includes: Constructing the network structures of the generator and discriminator, initializing the weight parameters and setting the hyperparameters; Extracting samples from the several corresponding retouching parameters screened out to form a training set and performing normalization preprocessing; During the iterative training, alternately fixing the discriminator and the generator, and inputting the new self-taken image to update the parameters of both; Based on the changes in the losses of the generator and discriminator, dynamically adjusting the hyperparameters and evaluating the quality of the new self-taken image; After the training is completed, importing the several corresponding retouching parameters screened out into the generator to obtain the beauty optimization result and using it for the beauty retouching of the new self-taken image.

6. An image processing system based on computer vision, characterized in that, Applying the computer vision-based image processing method according to any one of claims 1-5, the system includes: The acquisition module is used to acquire the historical satisfactory retouched images of the targeted user. The extraction module is used to extract the retouching parameters corresponding to the historical satisfactory retouched images. The first acquisition module is used to acquire a new self-taken image. The face detection module is used to perform face detection on the new self-taken image. The second acquisition module is used to acquire the key feature point position information of the face. The matching module is used to match the key feature point position information of the face in the new self-taken image with the retouching parameters corresponding to the historical satisfactory retouched images. The first generation module is used to generate the processed matching and screening results. The beauty retouching module is used to perform additional beauty retouching on the new self-taken image based on the processed matching and screening results. The second generation module is used to generate the additional beauty retouched image. The generative adversarial network module is used to optimize the retouching parameters corresponding to the historical satisfactory retouched images by the generative adversarial network during the process of performing additional beauty retouching on the new self-taken image.

7. The computer vision-based image processing system according to claim 6, wherein The matching module specifically includes: The dimensionality reduction processing unit is used to perform dimensionality reduction processing on the targeted feature vector. The calculation unit is used to calculate the similarity between the historical satisfactory retouched image and the feature vector after dimensionality reduction of the targeted feature vector using the cosine similarity algorithm. The verification unit is used to perform secondary verification on the cosine similarity calculation result by combining the Euclidean distance algorithm. The sorting unit is used to sort the retouching parameters corresponding to the historical satisfactory retouched images in descending order according to the similarity and distance calculation results. The screening unit is used to screen out several corresponding retouching parameters whose feature similarity with the new self-taken image is greater than the similarity threshold.

8. The image processing system based on computer vision according to claim 7, characterized in that, The beauty retouching module specifically includes: The determination unit is used to determine the initial parameter settings of the new image beauty algorithm model based on the several corresponding retouching parameters screened out. The fusion unit is used to fuse the determined initial parameter settings into the new self-taken image based on the image fusion algorithm based on the attention mechanism. The real-time monitoring unit is used to real-time monitor the color balance and contrast changes of the new self-taken image. The acquisition unit is used to acquire the adjustment instructions input by the targeted user in real time. The dynamic interaction unit is used to dynamically interact and adjust the parameter settings of the beauty algorithm model.

Images