Online intelligent appointment shooting system for shooting

Abstract

The invention relates to the field of intelligent appointment shooting, and discloses an online intelligent appointment shooting system for shooting, and the system comprises the steps: obtaining a background environment image of a target user; associating the background environment graph of the target user with the target work graph to obtain an associated work graph; pushing the associated work picture to the target user, and obtaining first behavior data of the target user on the associated work picture; analyzing the first behavior data of the target user, and determining an associated work diagram of the preference of the target user; and a style feature graph appointment shooting module of the target user is generated according to the related work graph of the preference of the target user and is used for acquiring second behavior data of the target user on the style feature graph, analyzing the second behavior data and determining a photographer pushing the corresponding related work graph to the target user.

Description

Technical Field

[0001] The present invention relates to the field of intelligent photography, and more specifically, to an online intelligent photography system for photography. Background Art

[0002] The photography appointment system is an online service model that has gradually emerged in the photography industry in recent years. It aims to connect photographers and users through digital means and provide users with a convenient photography appointment experience.

[0003] At present, most photo-shooting systems push photographers’ works to target users by randomly pushing photographers’ works or recommendation algorithms based on user interests. At the same time, they use Photoshop technology to generate virtual art photos to attract users’ interest in specific styles, thereby guiding users to accept offline photography invitations. However, this Photoshop-based method is usually difficult to reflect the real shooting effect due to the lack of integration with the user’s actual scene. Moreover, the lack of authenticity in the display method not only makes it difficult to stimulate users’ real interest, but also weakens users’ trust in the photographer’s works and services. Summary of the invention

[0004] The present invention provides an online intelligent photography appointment system for photographing, which solves the technical problems raised in the background technology.

[0005] The present invention provides an online intelligent photography appointment system for photography, comprising:

[0006] A data collection module is used to obtain the background environment map of the target user;

[0007] The image matching module is used to associate the background environment image of the target user with the target artwork image to obtain an associated artwork image;

[0008] A work push module is used to push the related work images to the target user and obtain the first action data of the target user on the related work images;

[0009] A style preference module is used to analyze the first behavior data of the target user and determine the target user's preferred related artwork map;

[0010] The image capture module is used to generate a style feature map of the target user based on the associated work map of the target user's preference, specifically including:

[0011] Embedding a texture description map into a computer screen of a target user to obtain a reference display of the computer screen; the texture description map represents a shooting posture of the target user;

[0012] Obtaining a first style feature vector of the preferred associated artwork image, dividing the reference display into a plurality of sub-areas according to the first style feature vector, and assigning different illumination intensities and illumination colors to each sub-area;

[0013] The photo shoot module is used to obtain the second behavior data of the target user for the style feature map, analyze the second behavior data, and determine the photographer who pushes the corresponding related work map to the target user.

[0014] Furthermore, the background environment map of the target user is obtained, including:

[0015] Step 1, collecting environmental images through the computer camera of the target user;

[0016] Step 2, semantically segmenting the environment image to obtain the human portrait outline in the environment image, and deleting the human portrait in the environment image according to the human portrait outline to obtain a segmentation map;

[0017] Step 3: fill the segmentation map with the padding model to obtain the background environment map.

[0018] Furthermore, the background environment map of the target user and the target artwork map are associated, including:

[0019] Step 4, generating a target background image for the target artwork image through steps 1 to 3;

[0020] Step 5, generating a texture feature matrix and a regional distribution matrix for the target background image and the background environment image respectively, specifically including:

[0021] Gray-scale the target background image and the background environment image to obtain a gray-scale target background image and a gray-scale background environment image; divide the gray-scale target background image and the gray-scale background environment image into m×n sub-blocks according to m rows and n columns, and calculate the average gradient strength and directional standard deviation for each sub-block; the texture feature matrix represents a matrix of size m×n, wherein the i-th row and j-th column of the texture feature matrix is ​​a texture vector, and the texture vector includes the average gradient strength and directional standard deviation of the corresponding sub-block;

[0022] The target background image and the background environment image are both semantically segmented by a preset semantic model to obtain a regional distribution matrix, which includes several elements; each element is assigned a different identification symbol;

[0023] Step 6: Calculate the correlation between the background environment image of the target user and the target artwork image based on the texture feature matrix and the regional distribution matrix. The correlation calculation formula is as follows: ;

[0024] in, Indicates the correlation between the target user's background environment map and the target artwork map. Represents the texture vector of the i-th row and j-th column in the texture feature matrix of the background environment image. Represents the texture vector of the i-th row and j-th column in the texture feature matrix of the target work image, represents the attenuation coefficient, represents the exponential function, , represents the natural base, Represents the operation of taking the absolute value. Represents the total number of background elements and non-background elements. The index representing the total number of categories. Represents the regional distribution matrix of the background environment map The coordinates of the center of the element, Represents the region distribution matrix of the target artwork. The coordinates of the center of the element, Indicates The importance weight of the elements, Represents the regional distribution matrix of the background environment map The regional distribution matrix of the element and the target artwork The consistency of the elements is 1 if they are the same, otherwise 0. represents the main weight parameter, represents the secondary weight parameter, and All are not 0, ;

[0025] Step 7, setting a correlation threshold. If the correlation between the background environment map of the target user and the target artwork map is greater than the correlation threshold, the target artwork map is correlated with the background environment map of the target user, otherwise, they are not correlated.

[0026] Furthermore, the first behavior data of the target user is analyzed to determine the target user's preferred related works graph, including:

[0027] Step 8, assigning different weights to each behavior in the first behavior data of the associated target artwork image, and calculating a weighted score for each target artwork image; wherein the first behavior data includes likes, favorites, stay time, and comments;

[0028] Step 9: The target artwork images whose weighted scores are greater than a preset score threshold are used as the target user's preferred associated artwork images.

[0029] Furthermore, the style feature vector of the preferred associated artwork graph is obtained, including:

[0030] Step 10, performing feature segmentation on the face part of the preferred associated work image to obtain key facial regions; the key facial regions include the forehead, eyes, nose, mouth and face;

[0031] Step 11, extracting the illumination intensity, illumination direction and illumination color for each key facial region; the illumination intensity is represented by the average value of the pixel brightness in the key facial region; the illumination direction is represented by the illumination direction obtained by performing light and shadow boundary detection on the key facial region; the illumination color is represented by the average color value in the extracted region;

[0032] Step 12: extracting the illumination intensity, illumination direction and illumination color according to each key facial region, and constructing a first style feature vector of the corresponding key facial region.

[0033] Furthermore, embedding a texture description map into the target user's computer screen includes:

[0034] Step 13, determining the shape and angle of each key facial area in the face part of the preferred associated artwork;

[0035] Step 14, obtaining the characteristic facial regions of the target user, the characteristic facial regions including the forehead, eyes, nose, mouth and face; fitting the characteristic facial regions with the key facial regions according to the shapes and angles of the key facial regions to obtain a texture description map; displaying the texture description map in the center of the computer screen to obtain a reference display of the computer screen.

[0036] Furthermore, the reference display is divided into a plurality of sub-areas according to the style feature vector, and different illumination intensities and illumination colors are assigned to each sub-area, including:

[0037] Step 15, randomly generate a population consisting of R individuals, and initialize the current iteration number to 1; encode each individual in the population, and the encoding is represented as an individual matrix with the same size as the matrix constructed by the pixels of the computer screen; divide the individual matrix into several sub-matrices according to the sub-areas of the computer screen, and the elements in each sub-matrix are the same. The elements in the individual matrix are represented as ;in, Expressed as light intensity, , and The color components are represented as red, green and blue respectively; , and are all non-negative numbers, , ;

[0038] Step 16, according to steps 10 to 12, obtaining a second style feature vector of a characteristic facial region of the target user in the style feature map of the target user obtained based on the individual matrix;

[0039] Step 17: construct an objective function based on the first style feature component and the second style feature component. The objective function is as follows: ;

[0040] in, Indicates The objective function value of each individual, and Represent the light intensity in the first style vector and the light intensity in the second eigenvector, respectively. represents the square of the Euclidean distance, and represent the lighting direction in the first style vector and the lighting direction in the second eigenvector, respectively. represents the vector transpose, represents the Frobenius norm squared, and Represent the lighting color in the first style vector and the lighting color in the second eigenvector, respectively. represents the L1 norm square;

[0041] Step 18, obtaining an objective function value for each individual according to the objective function, and sorting the individuals from small to large according to the objective function value to obtain a feature sorting; retaining a preset number of individuals from the front to the back of the feature sorting, and updating the remaining individuals as parents;

[0042] Step 19: If the current number of iterations reaches the preset number of iterations, the individual matrix of the individual with the smallest objective function value is output, otherwise return to step 18.

[0043] Furthermore, the remaining individuals are updated as parents, and the update formula is as follows: ;

[0044] in, Indicates the parent The individual matrix of The updated individual matrix is Indicates the parent The individual matrix of The individual matrix at Indicates the parent The individual matrix of The individual matrix at Indicates the number of iterations The individual matrix of the individual with the smallest global objective function value when , represents the first weight coefficient, represents the second weight coefficient, represents the third weight coefficient, Indicates the number of iterations The individual matrix and parent of the individual with the smallest global objective function value when The sum of the squared differences of the diagonal elements of the individual matrices of Indicates the preset number of iterations, Represents a random number, , Indicates the parent and parent The Hamming distance of Represents the square of the Euclidean norm.

[0045] Further, the second behavior data is analyzed to determine the photographer who pushes the corresponding related works to the target user, including:

[0046] Step 20, assigning different weights to each behavior in the second behavior data, and calculating a comprehensive score of the style feature graph; wherein the second behavior data includes likes, collections, stay time, and comments;

[0047] Step 21: If it is determined that the comprehensive score is greater than a preset comprehensive threshold, the photographer of the corresponding related works is pushed to the target user, otherwise the photographer of the corresponding related works is not pushed to the target user.

[0048] An online intelligent photo-shooting method for shooting, applied to the system, comprises:

[0049] Step 101, obtaining a background environment map of a target user;

[0050] Step 102, associating the background environment image of the target user with the target artwork image to obtain an associated artwork image;

[0051] Step 103, pushing the associated artwork map to the target user, and obtaining the target user's first behavior data on the associated artwork map;

[0052] Step 104, analyzing the first behavior data of the target user to determine a map of related works preferred by the target user;

[0053] Step 105, generating a style feature map of the target user according to the map of related works preferred by the target user, specifically includes:

[0054] Embedding a texture description map into a computer screen of a target user to obtain a reference display of the computer screen; the texture description map represents a shooting posture of the target user;

[0055] Obtaining a first style feature vector of the preferred associated artwork image, dividing the reference display into a plurality of sub-areas according to the first style feature vector, and assigning different illumination intensities and illumination colors to each sub-area;

[0056] Step 106, obtaining the second behavior data of the target user for the style feature graph, and analyzing the second behavior data to determine the photographer who should push the corresponding related work graph to the target user.

[0057] The beneficial effect of the present invention is that through the association matching and push mechanism based on the real background of the target user, combined with the dynamically generated style feature map and texture description map, the online photo-shooting process is highly authentic and accurate. Users can experience the style of the target work in a context that is highly consistent with their actual environment, and intuitively feel the final shooting effect, thereby effectively enhancing the user's sense of trust and participation, and improving the accuracy and efficiency of photo-shooting decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a module diagram of an online intelligent photo-shooting system for photo-shooting of the present invention;

[0059] Figure 2 It is a flow chart of an online intelligent photo-shooting method for photo-shooting according to the present invention. DETAILED DESCRIPTION

[0060] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the contents of this specification. Each example may omit, replace or add various processes or components as needed. In addition, the features described relative to some examples may also be combined in other examples.

[0061] like Figure 1-2 As shown, an online intelligent photo-shooting system for photo-shooting includes:

[0062] A data collection module is used to obtain the background environment map of the target user;

[0063] The image matching module is used to associate the background environment image of the target user with the target artwork image to obtain an associated artwork image;

[0064] A work push module is used to push the related work images to the target user and obtain the first action data of the target user on the related work images;

[0065] A style preference module is used to analyze the first behavior data of the target user and determine the target user's preferred related artwork map;

[0066] The image capture module is used to generate a style feature map of the target user based on the associated work map of the target user's preference, specifically including:

[0067] Embedding a texture description map into a computer screen of a target user to obtain a reference display of the computer screen; the texture description map represents a shooting posture of the target user;

[0068] Obtaining a first style feature vector of the preferred associated artwork image, dividing the reference display into a plurality of sub-areas according to the first style feature vector, and assigning different illumination intensities and illumination colors to each sub-area;

[0069] The photo shoot module is used to obtain the second behavior data of the target user for the style feature map, analyze the second behavior data, and determine the photographer who pushes the corresponding related work map to the target user.

[0070] In one embodiment of the present invention, obtaining a background environment map of a target user includes:

[0071] Step 1, collecting environmental images through the computer camera of the target user;

[0072] Step 2, semantically segmenting the environment image to obtain the human portrait outline in the environment image, and deleting the human portrait in the environment image according to the human portrait outline to obtain a segmentation map;

[0073] Step 3: fill the segmentation map with the padding model to obtain the background environment map.

[0074] Specifically, when the target user uses the system on a computer, the system obtains an external environment image through the computer's camera device, performs semantic segmentation on the environment image through the existing portrait recognition model, separates the portrait from the environment image, and obtains a segmentation map. Then, the missing part of the portrait in the segmentation map is supplemented through the existing image generation model to obtain a background environment map. Among them, the image generation model is supplemented by the non-portrait environment part in the segmentation map, so the obtained background environment map conforms to the real background of the target user.

[0075] In one embodiment of the present invention, associating the background environment image of the target user with the target artwork image includes:

[0076] Step 4, generating a target background image for the target artwork image through steps 1 to 3;

[0077] Step 5, generating a texture feature matrix and a regional distribution matrix for the target background image and the background environment image respectively, specifically including:

[0078] Gray-scale the target background image and the background environment image to obtain a gray-scale target background image and a gray-scale background environment image; divide the gray-scale target background image and the gray-scale background environment image into m×n sub-blocks according to m rows and n columns, and calculate the average gradient strength and directional standard deviation for each sub-block; the texture feature matrix represents a matrix of size m×n, wherein the i-th row and j-th column of the texture feature matrix is ​​a texture vector, and the texture vector includes the average gradient strength and directional standard deviation of the corresponding sub-block;

[0079] The target background image and the background environment image are both semantically segmented by a preset semantic model to obtain a regional distribution matrix, which includes several elements; each element is assigned a different identification symbol;

[0080] Step 6: Calculate the correlation between the background environment image of the target user and the target artwork image based on the texture feature matrix and the regional distribution matrix. The correlation calculation formula is as follows: ;

[0081] in, Indicates the correlation between the target user's background environment map and the target artwork map. Represents the texture vector of the i-th row and j-th column in the texture feature matrix of the background environment image. Represents the texture vector of the i-th row and j-th column in the texture feature matrix of the target work image, represents the attenuation coefficient, represents the exponential function, , represents the natural base, Represents the operation of taking the absolute value. Represents the total number of background elements and non-background elements. The index representing the total number of categories. Represents the regional distribution matrix of the background environment map The coordinates of the center of the element, Represents the region distribution matrix of the target artwork. The coordinates of the center of the element, Indicates The importance weight of the elements, Represents the regional distribution matrix of the background environment map The regional distribution matrix of the element and the target artwork The consistency of the elements is 1 if they are the same, otherwise 0. represents the main weight parameter, represents the secondary weight parameter, and All are not 0, ;

[0082] Step 7, setting a correlation threshold. If the correlation between the background environment map of the target user and the target artwork map is greater than the correlation threshold, the target artwork map is correlated with the background environment map of the target user, otherwise, they are not correlated.

[0083] Specifically, obtain the photographic works of the photographer registered in this system as the target work image. Perform the same operations as steps 1 to 3 on the target work image to obtain the target background of the target work image. Therefore, the matching correlation is calculated by the background environment image of the target user and the target background. If the correlation degree 0.9 is greater than the preset correlation degree threshold 0.7, then it can be considered that the target background of the target work image and the background environment image of the target user are the same. If the correlation degree 0.6 is less than the preset correlation degree threshold 0.7, then it can be considered that the target background of the target work image and the background environment image of the target user are different.

[0084] In one embodiment of the present invention, analyzing the first behavior data of the target user to determine the target user's preferred associated artwork graph includes:

[0085] Step 8, assigning different weights to each behavior in the first behavior data of the associated target artwork image, and calculating a weighted score for each target artwork image; wherein the first behavior data includes likes, favorites, stay time, and comments;

[0086] Step 9: The target artwork images whose weighted scores are greater than a preset score threshold are used as the target user's preferred associated artwork images.

[0087] Specifically, the target artwork images based on the target user's background association are used as associated artwork images, and the associated artwork images are pushed to the target user for browsing through the recommendation algorithm, and the first behavior data of the target user for all associated artwork images are collected. For example, the first behavior data of an associated artwork image is like, favorite, and stay time of 10 seconds, the weight assigned to like is 10, the weight assigned to favorite is 20, and the weight assigned to stay time is 2, then the weighted score of the corresponding associated artwork image is 1×10+1×20+10×2=50. If the preset score threshold is 30 points, then the corresponding associated artwork image is the preference of the target user.

[0088] In one embodiment of the present invention, obtaining a style feature vector of a preferred associated artwork image includes:

[0089] Step 10, performing feature segmentation on the face part of the preferred associated work image to obtain key facial regions; the key facial regions include the forehead, eyes, nose, mouth and face;

[0090] Step 11, extracting the illumination intensity, illumination direction and illumination color for each key facial region; the illumination intensity is represented by the average value of the pixel brightness in the key facial region; the illumination direction is represented by the illumination direction obtained by performing light and shadow boundary detection on the key facial region; the illumination color is represented by the average color value in the extracted region;

[0091] Step 12: extracting the illumination intensity, illumination direction and illumination color according to each key facial region, and constructing a first style feature vector of the corresponding key facial region.

[0092] Specifically, the facial part in the user's preferred associated work picture is semantically segmented and extracted through a portrait recognition model, and then the forehead, eyes, nose, mouth and face of the facial part are secondary segmented to obtain multiple key facial areas; the light intensity, light direction and light color are extracted for each key facial area, and then the first style feature vector of the facial part in the preferred associated work picture is obtained.

[0093] Specifically, the light intensity represents the brightness level of a key facial area, reflecting the degree of light in the area. Extraction method: First, the image of the target face is divided into several key facial areas, such as the forehead, eyes, nose, mouth and facial contour areas. In these divided areas, the brightness value of each pixel is read pixel by pixel. The brightness value is obtained by weighted calculation of the color information (red, green, blue) of the pixel. Then, the brightness values ​​of all pixels in the area are added and averaged to finally obtain the overall light intensity of the area. The extraction of light intensity can help determine the intensity of light received by a certain area of ​​the face, which is convenient for adjusting the light distribution in subsequent simulations, making the light effect more realistic and in line with the style of the target work. The light direction describes the angle of illumination of the light source, that is, from which direction the light is projected to the facial area. Extraction method: First, the image of the key facial area is subjected to light and shadow boundary detection. By detecting the transition boundary between light and dark in the facial area, it can be determined from which direction the light is irradiated to the face. Boundary detection can be completed by analyzing the trend of pixel brightness changes. The place with large brightness changes is usually the dividing line between light and shadow. After obtaining the light and shadow boundaries, calculate the normal direction of these boundaries. The direction of the normal can reflect the incident direction of the light. Finally, according to the law of the direction of light, the specific direction of the light source can be determined, such as from the upper left side, the front right side or other angles. The extraction of the light direction can help identify the position of the light source, so that the lighting angle can be adjusted in the subsequent light effect simulation, so that the light and shadow direction of the target user's photo is consistent with that of the target work. Light color is the color effect of light on the face, which indicates the overall color of the facial area under light. Extraction method: First, read the color value of each pixel in the key facial area, including three components of red, green and blue. Then, average the values ​​of the red, green and blue components respectively to obtain the average red value, green value and blue value of the area. The extraction of light color can help identify the color modification effect of light on the facial area. For example, a warm light source may make the facial area appear yellow, and a cold light source may make the face appear blue. This is particularly important for subsequent light effect adjustment and style imitation.

[0094] In one embodiment of the present invention, embedding a texture description map into a computer screen of a target user includes:

[0095] Step 13, determining the shape and angle of each key facial area in the face part of the preferred associated artwork;

[0096] Step 14, obtaining the characteristic facial regions of the target user, the characteristic facial regions including the forehead, eyes, nose, mouth and face; fitting the characteristic facial regions with the key facial regions according to the shapes and angles of the key facial regions to obtain a texture description map; displaying the texture description map in the center of the computer screen to obtain a reference display of the computer screen.

[0097] Specifically, it is necessary to illuminate the target user through the partition display of the computer screen, and then simulate the target user's preferred related work map to generate the corresponding style feature map. However, since the target user needs to remain delicate and the expression also needs to correspond to the portrait part in the preferred related work map, a texture feature map is constructed for the target user, and the texture feature map is embedded in the center of the computer screen through hidden pattern technology for display. When the target user simulates the texture feature map, the brightness of the matching part of the texture feature map is less than the part of the unmatched texture feature map. For example, if the target user's mouth matches the mouth of the texture feature map, the mouth of the texture feature map becomes darker, but the other parts maintain the original brightness, thereby controlling the target user to adjust his head position, facial expression and facial direction. It should be noted that the generation of the texture description map is based on the existing raw image model.

[0098] In one embodiment of the present invention, fitting the characteristic facial region to the key facial region includes:

[0099] The training data is constructed by manually drawing artificial texture description maps by painters, including: the shape, angle and expression features of the characteristic facial area (such as the curvature of the eyes, the upward corners of the mouth, etc.) and the key facial areas one by one. Synthetic training samples include pairing training data, using the artificial texture description map as label data, and the characteristic facial area of ​​the target user as the input sample. Each pair of training samples includes the characteristic facial area of ​​the target user and the artificial texture description map. Data enhancement includes increasing data diversity, rotating, scaling, cropping and other transformations on the celebrity texture description map and the target user image, and simulating different shooting angles and expression changes. The model network architecture uses the U-Net architecture (which is good at processing fine-grained changes in regional features). The input layer receives the characteristic facial area and key facial area of ​​the target user. The output layer generates the same artificial texture description as the celebrity. Figure 1The identity feature preservation module adds identity preservation loss to the network to prevent the generated result from losing the user's personal characteristics. The identity preservation loss loss function includes a combination of multiple loss functions. Pixel-level loss (L1 / L2 Loss): Maintain the consistency of the generated texture with the manually drawn texture. Feature matching loss: Use a pre-trained face model (such as VGG-Face) to extract features and maintain the user's identity. Model training, data partitioning, divide the training set, validation set, and test set according to 8:1:1. Ensure that the data in each segmented area is evenly distributed. During the training process, initialize the network weights and use the optimizer for optimization. Each training includes the following steps: Input the non-artificial texture description map of the target user. The model outputs the fitted texture description map. Calculate the loss and update the weights. The model weights that perform best on the validation set are saved for subsequent reasoning. Synthesize the generated results, recombine the fitted areas, and generate a complete target user's face image. Output the final image and display the generated image for the user to adjust and verify. Implement tools and frameworks. The facial feature extraction tool can be OpenCV, Dlib, or Mediapipe. The artificial texture drawing tool can be Procreate or Photoshop. The deep learning framework tool can be PyTorch or TensorFlow. The result evaluation uses SSIM and PSNR to evaluate the similarity between the generated image and the label.

[0100] In one embodiment of the present invention, the reference display is divided into a plurality of sub-areas according to the style feature vector, and different illumination intensities and illumination colors are assigned to each sub-area, including:

[0101] Step 15, randomly generate a population consisting of R individuals, and initialize the current iteration number to 1; encode each individual in the population, and the encoding is represented as an individual matrix with the same size as the matrix constructed by the pixels of the computer screen; divide the individual matrix into several sub-matrices according to the sub-areas of the computer screen, and the elements in each sub-matrix are the same. The elements in the individual matrix are represented as ;in, Expressed as light intensity, , and The color components are represented as red, green and blue respectively; , and are all non-negative numbers, , ;

[0102] Step 16, according to steps 10 to 12, obtaining a second style feature vector of a characteristic facial region of the target user in the style feature map of the target user obtained based on the individual matrix;

[0103] Step 17: construct an objective function based on the first style feature component and the second style feature component. The objective function is as follows: ;

[0104] in, Indicates The objective function value of each individual, and Represent the light intensity in the first style vector and the light intensity in the second eigenvector, respectively. represents the square of the Euclidean distance, and represent the lighting direction in the first style vector and the lighting direction in the second eigenvector, respectively. represents the vector transpose, represents the Frobenius norm squared, and Represent the lighting color in the first style vector and the lighting color in the second eigenvector, respectively. represents the L1 norm square;

[0105] Step 18, obtaining an objective function value for each individual according to the objective function, and sorting the individuals from small to large according to the objective function value to obtain a feature sorting; retaining a preset number of individuals from the front to the back of the feature sorting, and updating the remaining individuals as parents;

[0106] Step 19: If the current number of iterations reaches the preset number of iterations, the individual matrix of the individual with the smallest objective function value is output, otherwise return to step 18.

[0107] Specifically, when the target user completes the fitting with the texture feature map, the computer screen is divided into multiple light zones, each of which displays different brightness and color, and then the target user's face is illuminated to simulate the face in the preferred associated work picture. Several schemes are generated by the optimization algorithm, each of which should be applied to the target user at high frequency and for a short time, and the corresponding style feature map is obtained. The second eigenvector is obtained according to the style feature map, and the difference between the second eigenvector and the first eigenvector is calculated based on the light intensity, light direction and light color, and the difference is used as the objective function value. The smaller the objective function value, the more similar the style feature map and the corresponding preferred associated feature map are. Through iterative updating, after a preset number of iterations, the individual matrix of the individual with the smallest objective function value is obtained, and the light intensity, red color component, green color component and blue color component of each element of the individual matrix are configured with the corresponding computer screen pixel. Then the style feature map of the associated work picture that best meets the target user's preference is obtained. At the same time, excessive P-picture is avoided, and the realism of the style feature map is increased.

[0108] In one embodiment of the present invention, the remaining individuals are updated as parents, and the update formula is as follows: ;

[0109] in, Indicates the parent The individual matrix of The updated individual matrix is Indicates the parent The individual matrix of The individual matrix at Indicates the parent The individual matrix of The individual matrix at Indicates the number of iterations The individual matrix of the individual with the smallest global objective function value when , represents the first weight coefficient, represents the second weight coefficient, represents the third weight coefficient, Indicates the number of iterations The individual matrix and parent of the individual with the smallest global objective function value when The sum of the squared differences of the diagonal elements of the individual matrices of Indicates the preset number of iterations, Represents a random number, , Indicates the parent and parent The Hamming distance of Represents the square of the Euclidean norm.

[0110] In one embodiment of the present invention, analyzing the second behavior data to determine the photographer who pushes the corresponding related work image to the target user includes:

[0111] Step 20, assigning different weights to each behavior in the second behavior data, and calculating a comprehensive score of the style feature graph; wherein the second behavior data includes likes, collections, stay time, and comments;

[0112] Step 21: If it is determined that the comprehensive score is greater than a preset comprehensive threshold, the photographer of the corresponding related works is pushed to the target user, otherwise the photographer of the corresponding related works is not pushed to the target user.

[0113] Specifically, the second behavior data of the target user on the generated style feature map is obtained, and a comprehensive score for the style feature map is generated based on the second behavior data. If the comprehensive score is higher than a preset comprehensive threshold, the preferred photographer of the associated work map is recommended to the target user.

[0114] An online intelligent photo-shooting method for shooting, applied to the system, comprises:

[0115] Step 101, obtaining a background environment map of a target user;

[0116] Step 102, associating the background environment image of the target user with the target artwork image to obtain an associated artwork image;

[0117] Step 103, pushing the associated artwork map to the target user, and obtaining the target user's first behavior data on the associated artwork map;

[0118] Step 104, analyzing the first behavior data of the target user to determine a map of related works preferred by the target user;

[0119] Step 105, generating a style feature map of the target user according to the map of related works preferred by the target user, specifically includes:

[0120] Embedding a texture description map into a computer screen of a target user to obtain a reference display of the computer screen; the texture description map represents a shooting posture of the target user;

[0121] Obtaining a first style feature vector of the preferred associated artwork image, dividing the reference display into a plurality of sub-areas according to the first style feature vector, and assigning different illumination intensities and illumination colors to each sub-area;

[0122] Step 106, obtaining the second behavior data of the target user for the style feature graph, and analyzing the second behavior data to determine the photographer who should push the corresponding related work graph to the target user.

[0123] In one embodiment of the present invention, in order to achieve efficient appointment management for photo shoots, the system also includes an appointment process module to ensure that the appointment operation between the user and the photographer is convenient and smooth. The specific implementation is as follows:

[0124] User selects photographer: User browses through the recommended photographer list or searches for a specific photographer directly. Each photographer page displays detailed information, including portfolio, service content, price list, available appointment time, etc.

[0125] Time selection and reservation confirmation: The user enters the reservation page of the target photographer and checks the available time slots. The system displays the photographer's available time in a calendar format and marks the applicable time slots for different services (such as single-person portraits, wedding photography, etc.). After the user selects the desired time, he submits the reservation application.

[0126] Two-way confirmation mechanism: After the reservation application is submitted, the photographer will receive a system notification and confirm it within a limited time. Users can check the reservation status (pending, confirmed, rejected, etc.) in the system.

[0127] Payment and order generation: After the photographer confirms the appointment, the system prompts the user to complete the payment. The system generates an appointment order, including appointment details, payment status and contact information.

[0128] Reminders and follow-up services: The system automatically sends notification reminders to users and photographers before booking (via SMS or in-app messages). Users can submit special requests (such as venue layout, specific props, etc.) through the order page.

[0129] Reservation modification and cancellation:

[0130] Both users and photographers can apply to modify or cancel reservations through the system within the agreed modification period. Modification records and cancellation records will be automatically synchronized to the personal centers of both parties to ensure consistent information.

[0131] In one embodiment of the present invention, in order to enhance the interactivity and sense of community of the system, the system also includes a social attribute module to help users better interact and share with photographers and other users. The specific implementation is as follows:

[0132] User dynamic sharing: Users can upload their photos, write reviews or share their shooting experiences on their personal homepages. The system supports users to like, comment and forward dynamic content to enhance community interactivity.

[0133] Photographer homepage interaction: Each photographer has an independent homepage that displays his / her personal introduction, photography portfolio, and user reviews. Users can ask questions in the message board area of ​​the photographer's homepage (such as consulting style suggestions, understanding the shooting process, etc.). Photographers can reply to messages through the system to establish a closer connection with users.

[0134] Friends system: Users can add other users as friends through the system, view their friends’ shooting dynamics and collections. Friends can recommend photographers or works to each other, and create private chat groups to exchange shooting experiences.

[0135] Hot topics and discussion areas: The system regularly launches hot topics (such as "wedding photography tips", "most beautiful scene recommendations", etc.), and users can participate in discussions or share content. Discussion areas are categorized by topic, and photographers can also join the discussion and share professional advice with users.

[0136] Activities and community rewards: The platform regularly holds online activities (such as "Best Photography Works Selection", "Photography Experience Sharing Essays", etc.). Users can earn points by participating in the activities, which can be exchanged for coupons or free services within the platform.

[0137] Evaluation and feedback mechanism: After the user completes the shooting, he can rate the photographer's service and write a review in the system. The system generates a photographer rating list based on the accumulated reviews, which serves as an important reference for users to choose photographers. Photographers can also reply to user reviews and interact with users.

[0138] The above describes an embodiment of the present embodiment, but the present embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present embodiment, ordinary technicians in this field can also make many forms, all of which are within the protection of the present embodiment.

Claims

1. An online intelligent photo-shooting system for photography, characterized in that: include: A data collection module is used to obtain the background environment map of the target user; The image matching module is used to associate the background environment image of the target user with the target artwork image to obtain an associated artwork image; A work push module is used to push the related work images to the target user and obtain the first action data of the target user on the related work images; A style preference module is used to analyze the first behavior data of the target user and determine the target user's preferred related artwork map; The image capture module is used to generate a style feature map of the target user based on the associated work map of the target user's preference, specifically including: Embedding a texture description map into a computer screen of a target user to obtain a reference display of the computer screen; the texture description map represents a shooting posture of the target user; Obtaining a first style feature vector of the preferred associated artwork image, dividing the reference display into a plurality of sub-areas according to the first style feature vector, and assigning different illumination intensities and illumination colors to each sub-area; The photo shoot module is used to obtain the second behavior data of the target user for the style feature map, analyze the second behavior data, and determine the photographer who pushes the corresponding related work map to the target user.

2. The online intelligent photo-shooting system according to claim 1, characterized in that: Get the background environment map of the target user, including: Step 1, collecting environmental images through the computer camera of the target user; Step 2, semantically segmenting the environment image to obtain the human portrait outline in the environment image, and deleting the human portrait in the environment image according to the human portrait outline to obtain a segmentation map; Step 3: fill the segmentation map with the padding model to obtain the background environment map.

3. The online intelligent photo-shooting system according to claim 2, characterized in that: Associate the target user's background environment image with the target artwork image, including: Step 4, generating a target background image for the target artwork image through steps 1 to 3; Step 5, generating a texture feature matrix and a regional distribution matrix for the target background image and the background environment image respectively, specifically including: Graying the target background image and the background environment image to obtain a grayed target background image and a grayed background environment image; dividing the grayed target background image and the grayed background environment image into m×n sub-blocks according to m rows and n columns, and calculating the average gradient strength and the directional standard deviation for each sub-block; the texture feature matrix represents a matrix of size m×n, wherein the i-th row and j-th column of the texture feature matrix is ​​a texture vector, and the texture vector includes the average gradient strength and the directional standard deviation of the corresponding sub-block; The target background image and the background environment image are both semantically segmented by a preset semantic model to obtain a regional distribution matrix, which includes several elements; each element is assigned a different identification symbol; Step 6: Calculate the correlation between the background environment image of the target user and the target artwork image based on the texture feature matrix and the regional distribution matrix. The correlation calculation formula is as follows: ； in, Indicates the correlation between the target user's background environment map and the target artwork map. Represents the texture vector of the i-th row and j-th column in the texture feature matrix of the background environment image. Represents the texture vector of the i-th row and j-th column in the texture feature matrix of the target work image, represents the attenuation coefficient, represents the exponential function, , represents the natural base, Represents the operation of taking the absolute value. Represents the total number of background elements and non-background elements. The index representing the total number of categories. The regional distribution matrix of the background environment map The coordinates of the center of the element, Represents the region distribution matrix of the target artwork. The coordinates of the center of the element, Indicates The importance weight of the elements, The regional distribution matrix of the background environment map The regional distribution matrix of the element and the target artwork The consistency of the elements is 1 if they are the same, otherwise 0. represents the main weight parameter, represents the secondary weight parameter, and All are not 0, ; Step 7, setting a correlation threshold. If the correlation between the background environment map of the target user and the target artwork map is greater than the correlation threshold, the target artwork map is correlated with the background environment map of the target user, otherwise, they are not correlated.

4. The online intelligent photo-shooting system according to claim 3, characterized in that: Analyze the target user's first behavior data to determine the target user's preferred related works, including: Step 8, assigning different weights to each behavior in the first behavior data of the associated target artwork image, and calculating a weighted score for each target artwork image; wherein the first behavior data includes likes, favorites, stay time, and comments; Step 9: The target artwork images whose weighted scores are greater than a preset score threshold are used as the target user's preferred associated artwork images.

5. The online intelligent photo-shooting system according to claim 4, characterized in that: Obtain the style feature vector of the preferred associated artwork graph, including: Step 10, performing feature segmentation on the face part of the preferred associated work image to obtain key facial regions; the key facial regions include the forehead, eyes, nose, mouth and face; Step 11, extracting the illumination intensity, illumination direction and illumination color for each key facial region; the illumination intensity is represented by the average value of the pixel brightness in the key facial region; the illumination direction is represented by the illumination direction obtained by performing light and shadow boundary detection on the key facial region; the illumination color is represented by the average color value in the extracted region; Step 12: extracting the illumination intensity, illumination direction and illumination color according to each key facial region, and constructing a first style feature vector of the corresponding key facial region.

6. The online intelligent photo-shooting system according to claim 5, characterized in that: Embed a texture description map into the target user's computer screen, including: Step 13, determining the shape and angle of each key facial area in the face part of the preferred associated artwork; Step 14, obtaining the characteristic facial regions of the target user, the characteristic facial regions including the forehead, eyes, nose, mouth and face; fitting the characteristic facial regions with the key facial regions according to the shapes and angles of the key facial regions to obtain a texture description map; displaying the texture description map in the center of the computer screen to obtain a reference display of the computer screen.

7. The online intelligent photo-shooting system according to claim 6, characterized in that: The reference display is divided into multiple sub-areas according to the style feature vector, and different lighting intensities and lighting colors are assigned to each sub-area, including: Step 15, randomly generate a population consisting of R individuals, and initialize the current iteration number to 1; encode each individual in the population, and the encoding is represented as an individual matrix with the same size as the matrix constructed by the pixels of the computer screen; divide the individual matrix into several sub-matrices according to the sub-areas of the computer screen, and the elements in each sub-matrix are the same. The elements in the individual matrix are represented as ;in, Expressed as light intensity, , and The color components are represented as red, green and blue respectively; , and are all non-negative numbers, , ; Step 16, according to steps 10 to 12, obtaining a second style feature vector of a characteristic facial region of the target user in the style feature map of the target user obtained based on the individual matrix; Step 17: construct an objective function based on the first style feature component and the second style feature component. The objective function is as follows: ； in, Indicates The objective function value of each individual, and Represent the light intensity in the first style vector and the light intensity in the second eigenvector, respectively. represents the square of the Euclidean distance, and represent the lighting direction in the first style vector and the lighting direction in the second eigenvector, respectively. represents the vector transpose, represents the Frobenius norm squared, and Represent the lighting color in the first style vector and the lighting color in the second eigenvector, respectively. represents the L1 norm square; Step 18, obtaining an objective function value for each individual according to the objective function, and sorting the individuals from small to large according to the objective function value to obtain a feature sorting; retaining a preset number of individuals from the front to the back of the feature sorting, and updating the remaining individuals as parents; Step 19: If the current number of iterations reaches the preset number of iterations, the individual matrix of the individual with the smallest objective function value is output; otherwise, return to step 18.

8. The online intelligent photo-shooting system according to claim 7, characterized in that: The remaining individuals are updated as parents, and the update formula is as follows: ； in, Indicates the parent The individual matrix of The updated individual matrix is Indicates the parent The individual matrix of The individual matrix at Indicates the parent The individual matrix of The individual matrix at Indicates the number of iterations The individual matrix of the individual with the smallest global objective function value when , represents the first weight coefficient, represents the second weight coefficient, represents the third weight coefficient, Indicates the number of iterations The individual matrix and parent of the individual with the smallest global objective function value when The sum of the squared differences of the diagonal elements of the individual matrices of Indicates the preset number of iterations, Represents a random number, , Indicates the parent and parent The Hamming distance of Represents the square of the Euclidean norm.

9. The online intelligent photo-shooting system according to claim 8, characterized in that: Analyze the second behavior data to determine the photographer whose related works should be pushed to the target user, including: Step 20, assigning different weights to each behavior in the second behavior data, and calculating a comprehensive score of the style feature graph; wherein the second behavior data includes likes, collections, stay time, and comments; Step 21: If it is determined that the comprehensive score is greater than a preset comprehensive threshold, the photographer of the corresponding related works is pushed to the target user, otherwise the photographer of the corresponding related works is not pushed to the target user.

10. An online intelligent photo-shooting method for shooting, characterized in that: The system as claimed in any one of claims 1 to 9 comprises: Step 101, obtaining a background environment map of a target user; Step 102, associating the background environment image of the target user with the target artwork image to obtain an associated artwork image; Step 103, pushing the associated artwork map to the target user, and obtaining the target user's first behavior data on the associated artwork map; Step 104, analyzing the first behavior data of the target user to determine a map of related works preferred by the target user; Step 105, generating a style feature map of the target user according to the map of related works preferred by the target user, specifically includes: Embedding a texture description map into a computer screen of a target user to obtain a reference display of the computer screen; the texture description map represents a shooting posture of the target user; Obtaining a first style feature vector of the preferred associated artwork image, dividing the reference display into a plurality of sub-areas according to the first style feature vector, and assigning different illumination intensities and illumination colors to each sub-area; Step 106, obtaining the second behavior data of the target user for the style feature graph, and analyzing the second behavior data to determine the photographer who should push the corresponding related work graph to the target user.

Images