Special effect differentiation processing system based on face recognition processing

Abstract

The invention discloses a special effect differentiation processing system based on face recognition processing, and relates to the technical field of image processing, a face recognition module processes an image in real time to obtain face data, and a face weight distribution module calculates a weight based on a contour position, an area and features of five sense organs; the special effect distribution module matches special effects of corresponding computing power for different faces according to weights, the similar expression stabilization module dynamically evaluates the stability of special effect generation through statistical analysis, intelligent distribution and optimization of special effect resources are achieved, the key face special effect is improved, and computing power consumption is reduced. Priorities are distributed for different faces through a face weight distribution module (position, area and feature weight of five sense organs), high computing power special effects are generated only for high-weight faces by combining computing power grading (special effect cell division) of a special effect distribution module, and low-weight faces are matched with lightweight or do not generate special effects.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and particularly to a special effect differentiation processing system based on face recognition processing. Background Art

[0002] Images can be greatly enhanced in visual attraction through special effects, attracting the audience's attention by creating fantasy scenes, exaggerating colors, etc.; it can help realize the unique creativity and artistic concept of creators, integrating reality and imagination and breaking through the limitations of real shooting; it can enhance the narrative ability of images, such as better conveying story information and emotions by adding specific elements or scene changes; it can also meet the specific needs of different industries, such as highlighting product features in advertisements, creating immersive virtual environments in games, and providing personalized expressions for users on social platforms, making images more interesting and communicable, thus playing an important role in information dissemination, culture and entertainment, etc.

[0003] Existing technologies usually first use deep learning algorithms for face detection and key point positioning to accurately identify the position of the face and key parts such as facial features, then extract the feature data of the face, and then, according to the preset special effect types, perform operations such as stretching and distorting the facial features through image deformation technology, or use texture mapping and image fusion technology to superimpose virtual textures, patterns, etc. on the face area, and will also use 3D modeling and animation technology to construct dynamic facial special effects, such as simulating expression changes such as blinking and smiling. Finally, the processed face image with special effects is quickly output through real-time rendering technology to achieve real-time display and interaction.

[0004] In the scenario where an image contains multiple faces, some faces are not the required main bodies. However, the current special effect system does not distinguish and assigns special effects to each face, which not only causes waste of special effect computing power resources, but also weakens the special effect effects when key figures make similar expressions, resulting in chaotic picture information and inability to highlight the core content, greatly affecting the expected presentation effect and practical application value of the special effects. Summary of the Invention

[0005] Technical Problem to be Solved

[0006] Aiming at the deficiencies of the existing technologies, the present invention provides a special effect differentiation processing system based on face recognition processing, which solves the problems of waste of computing power resources caused by the special effect system assigning special effects to non-main body faces in the multi-face image scenario and the inability to highlight the core when key figures make similar expressions.

[0007] Technical Solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: A special effect differentiation processing system based on face recognition processing, including the following specific modules: Face recognition module: Processes the real-time acquired image data to obtain face image data and sends it to the face weight allocation module; Face weight allocation module: Processes the face image data through the contour position weight algorithm, the contour area weight algorithm, and the facial feature weight algorithm to obtain the facial position weight, the facial area weight, and the facial feature weight and sends them to the special effect allocation module; Special effect allocation module: Allocates special effects according to the weights of the face; Similar expression stability module: Dynamically evaluates the stability of special effects during similar expressions. If it is lower than the threshold, it continuously generates special effects. Otherwise, it suspends generating special effects during similar expressions.

[0009] Further, in the face recognition module, the image data includes the number of pixel points, the pixel point coordinates, and the pixel point brightness value. The real-time acquired image data is subjected to filtering and noise reduction processing, and the filtered and noise-reduced image data is processed through the image contour segmentation algorithm and the image contour tracking algorithm to obtain the face image data.

[0010] Further, in the face weight allocation module, the facial position weight, the facial area weight, and the facial feature weight are summed to calculate the weight data of different faces. Among them, RQ represents the weight data of different faces, o represents the number of frames of the image, LW represents the facial position weight, LM represents the facial area weight, and WT represents the facial feature weight.

[0011] Further, the specific method for obtaining the facial position weight is as follows: Calculate the actual facial position distance by using the Euclidean distance formula for the image data and the face image data. Calculate the average of the actual facial positions during the time period from the moment when the face image data is obtained to the current moment to obtain the average facial position distance; Set the ordinary weight, sort the actual facial position distances of different faces in descending order by bubble sort, allocate n ordinary weights to the face with the largest actual facial position distance, allocate n - 1 ordinary weights to the face with the second largest actual facial position distance, and so on, allocate 1 ordinary weight to the face with the smallest actual facial position distance. Compare the actual facial position distance of each face with the average facial position distance. If the actual facial position distance is greater than the average facial position distance, delete 1 weight from this face; If the actual facial position distance is equal to the average facial position distance, do not operate on this face; If the actual facial position distance is less than the average facial position distance, add 1 weight to this face.

[0012] Further, the specific method for obtaining the actual face position distance is as follows: Sum up the pixel point coordinates in the human face and then calculate the average to obtain the center point of the human face. Sum up the pixel point coordinates in the image data and then calculate the average to obtain the center point of the image. Where SJ represents the actual face position, xi represents the abscissa of the center point of the human face, x represents the abscissa of the center point of the image, and y i represents the ordinate of the center point of the human face, and y represents the ordinate of the center point of the image.

[0013] Further, the specific method for obtaining the face area weight is as follows: Sum up the number of pixel points in different human faces in sequence to obtain the face areas respectively. Sort the different face areas in ascending order through bubble sort. Determine the lower limit and upper limit of the area range with the smallest face area and the largest face area. Divide the area range into n sub-ranges evenly according to the number of human faces. Sort the upper limits of the sub-ranges in ascending order through bubble sort, and limit the weights for each sub-range in sequence. The largest sub-range is limited with n weights, the second largest sub-range is limited with n - 1 weights, and so on. The smallest sub-range is limited with 1 weight. Compare the lower limit and upper limit of the sub-ranges in sequence from the smallest face area to the largest face area. If the face area is within the lower limit and upper limit of the corresponding sub-range, the sub-range assigns the limited weight to the face area. If the face area is not within the lower limit and upper limit of the corresponding sub-range, continue to traverse until it is within the lower limit and upper limit of other sub-ranges.

[0014] Further, the specific method for obtaining the facial feature weight is as follows: Process the face image data through the image contour segmentation algorithm and the image contour tracking algorithm to obtain the facial feature contour data. Process the facial feature contour data through the CNN model expression algorithm to obtain the facial expression features of the facial features. Convert each facial expression feature into a vector. Compare the expression similarities of two different facial expression features in n human faces respectively through the cosine similarity method. If the expressions of two different facial expression features are similar, assign special weights to these two different human faces. If the expressions of two different facial expression features are not similar, do not assign special weights to these two different human faces, and so on until all human faces are compared.

[0015] Further, in the special effect allocation module, weight data of different faces is received and special effects are divided and processed according to different consumption of computing power resources to obtain different special effect sub - intervals. The weights of the faces are compared in turn within the range from the lower limit to the upper limit of the special effect sub - intervals. If the weight of a face is within the range of a special effect sub - interval, the special effect of this special effect sub - interval is generated for this face. If the weight of the face is not within the range of the special effect sub - interval, continue to traverse until each special effect sub - interval is traversed.

[0016] Further, the specific method for obtaining the special effect sub - intervals is as follows: Each special effect is sorted in ascending order according to different computing power consumption through bubble sorting, and adjacent special effect computing powers are successively made to form continuous special effect sub - intervals; the lower limit of the smallest special effect sub - interval is denoted as XQ, and it forms a zero interval with zero, and the zero interval is continuous with the smallest special effect sub - interval. When the weight of a face is within the zero interval, no special effect is generated; a partition sub - interval is set, and the lower limit of this interval is not continuous with the upper limit of the largest special effect sub - interval, and the lower limit of this interval is much larger than the upper limit of the largest special effect sub - interval.

[0017] Further, in the similar expression stabilization module, within the time period from the moment when the face image data is obtained to the current moment, the time for generating special effects each time the expression of the same face is similar is recorded to obtain the similar special effect time. The average value of the similar special effect time is calculated to obtain the average similar special effect time. The standard deviation of the similar special effect time is calculated according to the average similar special effect time to obtain the similar special effect stability value. By comparing the similar special effect stability value with the stability threshold, if the similar special effect stability value is less than the stability threshold, continue to generate special effects when the expression of this face is similar in the future. If the similar special effect stability value is greater than or equal to the stability threshold, do not generate special effects when the expression of this face is similar for a short time.

[0018] Beneficial effects

[0019] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0020] 1. The face weight allocation module (position, area, facial feature weights) assigns priorities to different faces. Combining with the computing power grading (special effect sub - interval division) of the special effect allocation module, only high - weight faces generate high - computing - power special effects, and low - weight faces are matched with lightweight special effects or no special effects are generated.

[0021] 2. The facial feature weight algorithm compares the expression similarity through cosine similarity, and assigns special weights to key faces with similar expressions to ensure their special effect priorities; the special effect allocation module prevents non - similar expression faces from accidentally triggering high - computing - power special effects through the "partition sub - interval" design.

[0022] 3. The Similar Expression Stabilization Module records the historical special effect generation time, calculates the average interval and standard deviation, and dynamically adjusts the generation strategy to reduce the system latency caused by frequent state switches.

[0023] Of course, it is not necessary for any product implementing the present invention to achieve all the above-mentioned advantages simultaneously. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a structural diagram of a special effect differentiation processing system based on face recognition processing according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0026] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0027] As Figure 1 shown, the embodiments of the present invention provide a special effect differentiation processing system based on face recognition processing, including the following specific modules:

[0028] Face Recognition Module: Real-time obtains image data through a camera. The image data includes the number of pixel points, pixel point coordinates, and pixel point brightness values. Performs filtering and noise reduction processing on the image data, which helps to improve the quality of the image data. Processes the filtered and noise-reduced image data through an image contour segmentation algorithm and an image contour tracking algorithm to obtain face image data. There are n faces in the face image data, and the face image data is sent to the Face Weight Allocation Module;

[0029] Image contour segmentation algorithm, such as the Sobel algorithm, first converts the image data into a grayscale image to simplify the calculation; then uses 3×3 convolution kernels in the horizontal and vertical directions (the horizontal convolution kernel focuses on detecting vertical edges, and the vertical convolution kernel focuses on detecting horizontal edges) to perform convolution operations on the grayscale image to obtain a horizontal gradient matrix and a vertical gradient matrix; then performs square sum and square root calculations on the horizontal gradient matrix and the vertical gradient matrix to obtain the gradient magnitude, which is used to determine the edges, and uses the arctangent function to calculate to obtain the gradient direction, which is used to determine the direction in which the edges extend; finally, performs binary processing on the gradient magnitude by setting a threshold, and marks the pixel coordinates with a gradient magnitude higher than the threshold as the contour;

[0030] Image contour tracking algorithm, such as the Lucas-Kanade optical flow method, first makes an assumption of constant pixel brightness value to obtain an assumption formula, that is, the brightness values of the same object pixels in two adjacent frames of images are unchanged; then performs a first-order Taylor expansion on the assumption formula to obtain an optical flow constraint equation; then uses the Sobel algorithm and the difference between adjacent frames to calculate to obtain the spatial gradient and the temporal gradient, which are used to reflect the influence of the changes in the spatial gradient and the temporal gradient in the optical flow constraint equation on the pixel brightness value; since there are two unknowns in the optical flow constraint equation, it is assumed that the pixels in the neighborhood move the same, so the optical flow constraint equations of the pixels in the neighborhood are combined into an overdetermined system of equations; then uses the least squares method to solve the system of equations to avoid errors and obtain an estimated value of the optical flow velocity; finally, in practical applications, feature points (such as corner points) are selected, optical flow calculations and tracking are performed on them, and then the positions of the feature points are updated in each frame to achieve target tracking.

[0031] Face weight allocation module: Receives face image data and processes the face image data through the contour position weight algorithm, the contour area weight algorithm, and the facial features weight algorithm to obtain the facial position weight, the facial area weight, and the facial features weight in the face image data that changes in each frame, and performs a summation calculation on the facial position weight, the facial area weight, and the facial features weight to obtain the weight data of different faces, where RQ represents the weight data of different faces, o represents the number of frames of the image. Since the face changes in each frame of the image, the face weights in each frame of the image will also be different. LW represents the facial position weight, which reflects the magnitude of the facial position weight. LM represents the facial area weight, which reflects the magnitude of the facial area weight. WT represents the facial features weight, which reflects the magnitude of the facial features weight. The weight data of different faces is sent to the special effect allocation module.

[0032] The specific acquisition method of obtaining the facial position weight through the contour position weight algorithm is as follows:

[0033] Calculate the Euclidean distance formula for the image data and the face image data. The Euclidean distance formula is the formula for calculating the straight-line distance between the coordinates of two pixel points in a two-dimensional plane, and obtain the actual face position distance. The actual face position distance refers to the distance from the center point of the face to the center point of the image. The greater the actual face position distance of the face, the more marginalized the face is, and then the lower the weight. Calculate the average of the actual face positions during the time period from the moment when the face image data is obtained to the current moment, and obtain the average face position distance, which represents the average distance of the center point of each face to the center point of the image in different image frames;

[0034] Set a general weight. The general weight is a constant q greater than one. Arrange the actual face position distances of different faces in descending order through bubble sort. Bubble sort is to compare and exchange the sizes of adjacent weights, and gradually move the smallest weight to the end of the array, and finally achieve sorting from large to small. Assign n general weights to the face with the largest actual face position distance, assign n - 1 general weights to the face with the second largest actual face position distance, and so on. Assign 1 general weight to the face with the smallest actual face position distance. Compare the actual face position distance of each face with the average face position distance. If the actual face position distance is greater than the average face position distance, delete 1 weight from this face, indicating that the importance of this face decreases; if the actual face position distance is equal to the average face position distance, do not operate on this face, indicating that the importance of this face remains unchanged; if the actual face position distance is less than the average face position distance, add 1 weight to this face, indicating that the importance of this face increases.

[0035] The specific method for obtaining the actual face position distance is as follows:

[0036] Sum up and then take the average of the pixel point coordinates in the face to obtain the center point of the face, which is used as the position representation of the face in the image. Sum up and then take the average of the pixel point coordinates in the image data to obtain the center point of the image. where SJ represents the actual face position, x i represents the abscissa of the center point of the face, x represents the abscissa of the center point of the image, and y i represents the ordinate of the center point of the face, and y represents the ordinate of the center point of the image.

[0037] The specific method for obtaining the face area weight through the contour area weight algorithm is as follows:

[0038] Since the face image data is obtained from the image data, the face image data includes the number of pixel points, the pixel point coordinates, and the pixel point brightness values. Calculate the sum of the number of pixel points in different faces in sequence to obtain the face areas respectively. The larger the area of a face, the more important this face is, and thus the higher the weight. Sort the different face areas in ascending order through bubble sort, and determine the lower and upper limits of the area range with the smallest face area and the largest face area. The area range is used to limit the weight of the face area. Divide the area range into n sub-ranges evenly according to the number of faces. Sort the upper limits of the sub-ranges in ascending order through bubble sort, and limit the weights for each sub-range in sequence to allocate the weights corresponding to the face areas. The largest sub-range limits n weights, the second largest sub-range limits n - 1 weights, and so on. The smallest sub-range limits 1 weight;

[0039] Starting from the smallest face area to the largest face area, compare the lower and upper limits of the sub-ranges in sequence. If the face area is within the lower and upper limits of the corresponding sub-range, the sub-range allocates the already limited weight to the face area. If the face area is not within the lower and upper limits of the corresponding sub-range, continue to traverse until it is within the lower and upper limits of other sub-ranges.

[0040] The specific way to obtain the facial feature weights through the facial feature weight algorithm is as follows:

[0041] Process the face image data through the image contour segmentation algorithm and the image contour tracking algorithm to obtain the facial feature contour data. Since the facial feature contour data is obtained from the face image data, the facial feature contour data includes the number of pixel points, the pixel point coordinates, and the pixel point brightness values. Process the facial feature contour data through the CNN model expression algorithm. Specifically, extract the facial feature contours through the convolutional layer and learn according to the dynamic expression information of the facial feature contours, and remove the information irrelevant to this information, retaining the key features of the expression to obtain the facial feature expression features. Convert each facial feature expression feature into a vector, and compare the expression similarities of two different facial feature expression features in n faces respectively through the cosine similarity method;

[0042]

[0043] Among them, BX represents the expression similarity, reflecting the expression similarity when two different facial feature expression features rotate in three-dimensional space. Since the expression similarity is represented by cosine, the range is from negative one to positive one. SP represents the range of the face rotating horizontally from left to right, which is from negative ninety degrees to positive ninety degrees. CZ represents the range of the face tilting from down to up, which is from negative ninety degrees to positive ninety degrees. CF represents the angle when the human body tilts the head, with the left ear against the left shoulder being negative ninety degrees and the right ear against the right shoulder being positive ninety degrees;

[0044] If the expressions of two different facial feature expressions are similar, that is, the expression similarity is greater than zero and less than or equal to one, then special weights are assigned to these two different faces, and the special weight is the exponential e q , if the expressions of two different facial feature expressions are not similar, that is, the expression similarity is greater than or equal to negative one and less than or equal to zero, then no special weights are assigned to these two different faces, and so on until each face has been compared

[0045] Special effect allocation module: Receives the weight data of different faces and divides and processes the special effects according to different consumption of computing power resources to obtain different special effect sub-intervals. Compare the weights of the faces in turn within the range from the lower limit to the upper limit of the special effect sub-intervals. If the weight of the face is within the range of the special effect sub-interval, then generate the special effect of this special effect sub-interval for this face. If the weight of the face is not within the range of the special effect sub-interval, then continue to traverse until each special effect sub-interval has been traversed

[0046] The specific method for obtaining the special effect sub-intervals is as follows

[0047] Sort each special effect in ascending order according to different computing power consumption through bubble sort, and successively make the adjacent special effect computing powers form continuous special effect sub-intervals; the lower limit of the smallest special effect sub-interval is denoted as XQ, and combined with zero to form a zero interval. The lower limit of the zero interval is zero, and the upper limit is XQ, and the zero interval is continuous with the smallest special effect sub-interval. When the weight of the face is within the zero interval, no special effect is generated; set a partition sub-interval, the lower limit of this interval is not continuous with the upper limit of the largest special effect sub-interval, and the lower limit of this interval is much larger than the upper limit of the largest special effect sub-interval. Since the special weight is much larger than the ordinary weight, it is avoided that when the expressions of different faces are not similar, special effects are generated only through the facial position weight and the facial area weight when the expressions are similar

[0048] Similar expression stability module: During the time period from the moment when the face image data is obtained to the current moment, record the time when the special effect is generated each time the expression of the same face is similar to obtain the similar special effect time. Calculate the average value of the similar special effect time to obtain the average similar special effect time. Use the average similar special effect time as a measure of whether the similar special effect time is stable. Calculate the standard deviation of the similar special effect time based on the average similar special effect time to obtain the similar special effect stability value. Detect whether the similar special effect stability value is stable through the volatility of the similar special effect stability value on the average similar special effect time. Compare the similar special effect stability value with the stability threshold. If the similar special effect stability value is less than the stability threshold and the fluctuation of the similar special effect stability value is small, then continue to generate the special effect when the expression of this face is similar in the future. If the similar special effect stability value is greater than or equal to the stability threshold and the fluctuation of the similar special effect stability value is large, then do not generate the special effect when the expression of this face is similar in a short time, so as to avoid increasing the system overhead caused by frequent trigger of stop or resume operations and save computing power resources

[0049] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. A special effect differentiation processing system based on face recognition processing, characterized in that: It includes the following specific steps: Face recognition module: Processes the image data obtained in real time, obtains the face image data, and sends it to the face weight assignment module; Face weight assignment module: Processes the face image data through the contour position weight algorithm, the contour area weight algorithm, and the facial feature weight algorithm, obtains the facial position weight, the facial area weight, and the facial feature weight, and sends them to the special effect assignment module; Special effect assignment module: Assigns special effects according to the weights of the face; Similar expression stability module: Dynamically evaluates the stability of special effects during similar expressions. If it is lower than the threshold, it continuously generates special effects. Otherwise, it pauses generating special effects during similar expressions.

2. The special effect differentiation processing system based on face recognition processing according to claim 1, wherein: In the face recognition module, the image data includes the number of pixel points, the pixel point coordinates, and the pixel point brightness value. The image data obtained in real time is subjected to filtering and noise reduction processing, and the filtered and noise-reduced image data is processed through the image contour segmentation algorithm and the image contour tracking algorithm to obtain the face image data.

3. The special effect differentiation processing system based on face recognition processing according to claim 1, wherein: In the face weight distribution module, the face position weight, face area weight, and facial feature weights are summed up to obtain the weight data of different faces. Among them, RQ represents the weight data of different faces, o represents the number of frames of the image, LW represents the face position weight, LM represents the face area weight, and WT represents the facial feature weights.

4. The special effect differentiation processing system based on face recognition processing according to claim 3, characterized in that: The specific method for obtaining the facial position weight is as follows: Calculate the actual facial position distance by using the Euclidean distance formula for the image data and the face image data. Calculate the average of the actual facial positions within the time period from the moment when the face image data is obtained to the current moment to obtain the average facial position distance; Set the ordinary weights, sort the actual facial position distances of different faces in descending order by bubble sort, assign n ordinary weights to the face with the largest actual facial position distance, assign n - 1 ordinary weights to the face with the second largest actual facial position distance, and so on, assign 1 ordinary weight to the face with the smallest actual facial position distance. Compare the actual facial position distance of each face with the average facial position distance. If the actual facial position distance is greater than the average facial position distance, delete 1 weight from this face; If the actual facial position distance is equal to the average facial position distance, no operation is performed on this face; If the actual facial position distance is less than the average facial position distance, add 1 weight to this face.

5. The special effect differentiation processing system based on face recognition processing according to claim 4, characterized in that: The specific method for obtaining the actual facial position distance is as follows: Sum the pixel coordinates in the face and then calculate the average to obtain the center point of the face. Sum the pixel coordinates in the image data and then calculate the average to obtain the center point of the image. Where SJ represents the actual face position, x i represents the abscissa of the center point of the face, x represents the abscissa of the center point of the image, y i represents the ordinate of the center point of the face, y represents the ordinate of the center point of the image.

6. The special effect differentiation processing system based on face recognition processing according to claim 3, characterized in that: The specific method for obtaining the facial area weight is as follows: Sum up the number of pixel points in different faces in turn to obtain the face areas respectively; Sort the different face areas in ascending order by bubble sort, determine the lower and upper limits of the area interval with the smallest face area and the largest face area, divide the area interval into n sub-intervals on average according to the number of faces, sort the upper limits of the sub-intervals in ascending order by bubble sort, and limit the weights for each sub-interval in turn. The largest sub-interval is limited with n weights, the second largest sub-interval is limited with n - 1 weights, and so on, the smallest sub-interval is limited with 1 weight; Starting from the minimum face area to the maximum face area, compare the lower and upper limits of the sub-intervals in sequence. If the face area is within the lower and upper limits of the corresponding sub-interval, the sub-interval assigns the pre-defined weight to the face area. If the face area is not within the lower and upper limits of the corresponding sub-interval, continue to traverse until it is within the lower and upper limits of other sub-intervals.

7. An effect differentiation processing system based on face recognition processing according to claim 3, characterized in that: The specific method for obtaining the weights of the facial feature is as follows: Process the facial image data through the image contour segmentation algorithm and the image contour tracking algorithm to obtain the facial feature contour data. Process the facial feature contour data through the CNN model expression algorithm to obtain the facial feature expression features. Convert each facial feature expression feature into a vector, and use the cosine similarity method to compare the expression similarities of two different facial feature expression features among n faces respectively; If the expressions of two different facial feature expression features are similar, assign special weights to these two different faces. If the expressions of two different facial feature expression features are not similar, do not assign special weights to these two different faces, and so on until all faces are compared.

8. The special effect differentiation processing system based on face recognition processing according to claim 1, characterized in that: In the special effect allocation module, receive the weight data of different faces and divide the special effects according to different computing power resources consumed, obtaining different special effect sub-intervals. Compare the weights of the faces in sequence within the range from the lower limit to the upper limit of the special effect sub-intervals. If the weight of the face is within the range of the special effect sub-interval, generate the special effect of this special effect sub-interval for the face. If the weight of the face is not within the range of the special effect sub-interval, continue to traverse until all special effect sub-intervals are traversed.

9. The special effect differentiation processing system based on face recognition processing according to claim 8, wherein: The specific method for obtaining the special effect sub-intervals is as follows: Sort each special effect in ascending order according to different computing power consumption through bubble sort, and successively form continuous special effect sub-intervals with adjacent special effect computing powers; Denote the lower limit of the smallest special effect sub-interval as XQ, and form a zero interval by combining XQ with zero. The zero interval is continuous with the smallest special effect sub-interval. When the weight of the face is within the zero interval, no special effect is generated; Set a partition sub-interval, the lower limit of this interval is not continuous with the upper limit of the largest special effect sub-interval, and the lower limit of this interval is much larger than the upper limit of the largest special effect sub-interval.

10. The special effect differentiation processing system based on face recognition processing according to claim 1, characterized in that: In the similar expression stability module, within the time period from the moment when the facial image data is obtained to the current moment, record the time when the special effect is generated each time the expression of the same face is similar, obtaining the similar special effect time. Calculate the average value of the similar special effect time to obtain the average similar special effect time. Calculate the standard deviation of the similar special effect time based on the average similar special effect time to obtain the similar special effect stability value. Compare the similar special effect stability value with the stability threshold. If the similar special effect stability value is less than the stability threshold, continue to generate the special effect when the expression of the face is similar subsequently. If the similar special effect stability value is greater than or equal to the stability threshold, do not generate the special effect when the expression of the face is similar within a short period of time.

Images