Method and system for face rejuvenation of movie characters based on deep learning

Abstract

The present application relates to the technical field of image smoothing filtering, in particular to a film and television character face rejuvenation method and system based on deep learning. The method identifies wrinkle areas and eye areas in a film and television character face image, inputs the face image into a deep convolutional network, and the deep convolutional network extracts features of organ expression change parameters of the eye area and features of wrinkle appearance coefficients of the face image. Through analysis between adjacent frames, the wrinkle appearance correlation parameters of the face image are determined. The smallest wrinkle appearance coefficient is taken as a reference frame to obtain a wrinkle appearance elimination coefficient, and the wrinkle appearance correlation parameters are used to adjust the wrinkle appearance elimination parameters, so as to realize adaptive smoothing of the face image and obtain a face rejuvenation result. Through adjustment of the smoothing parameters of the character face image, the present application avoids overfitting and unnatural problems of the rejuvenation result, and improves the quality of the rejuvenation result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image smoothing filtering, in particular to a film and television character face rejuvenation method and system based on deep learning. BACKGROUND

[0002] Due to the pursuit of character shaping reality and diversity by the film industry, and the expectation of high-quality visual experience by the audience, along with the rapid development of computer graphics image science and computer hardware and software technology, the digital production mode of the film has completely replaced the film production mode. The emergence of digital technology not only greatly enriches the production means of the film, but also provides more possibilities for the shaping of character image, thereby realizing the face rejuvenation of film and television characters. The most important step of rejuvenation is to smooth the wrinkles on the face of the character.

[0003] Because there are differences in facial morphology and expression, and the degree of wrinkle appearance is different at different times, if the fixed smoothing parameter is directly used to smooth the wrinkles in the face image, it will cause obvious discomfort and affect the effect of character rejuvenation. SUMMARY

[0004] In order to solve the technical problem that the existing technology smoothes the wrinkles in the face image by using fixed smoothing parameters, affecting the effect of character rejuvenation, the purpose of the present application is to provide a film and television character face rejuvenation method and system based on deep learning, and the technical solution adopted is as follows:

[0005] The present application proposes a film and television character face rejuvenation method based on deep learning, which comprises:

[0006] Obtain multiple frames of face images of film and television characters, identify the wrinkle area and eye area in the face image, identify the edge contour and center point of the eye area; input the face image into a deep convolutional network to obtain the face rejuvenation result; the method for processing the face image by the deep convolutional network comprises:

[0007] Step S1: taking the line between the center point and each edge contour point as a reference line; obtaining the organ expression change parameter of the eye area according to the discontinuity of the edge contour point and the length of the reference line;

[0008] Step S2: extending the reference line, and taking the intersection point of the nearest wrinkle area as the associated wrinkle point; taking the line between the associated wrinkle point as the associated wrinkle line in each wrinkle area; obtaining the wrinkle appearance coefficient of each wrinkle area according to the proportion of the associated wrinkle line in the wrinkle area and the pixel value of the wrinkle area;

[0009] Step S3: Obtaining the overall wrinkle appearance coefficient of the current facial image based on the wrinkle appearance coefficients of all wrinkle areas, and obtaining the wrinkle appearance correlation parameters of each facial image frame based on the differences in the overall wrinkle appearance coefficients and the differences in organ expression change parameters between adjacent facial images;

[0010] Step S4: Among multiple frames of facial images, select the smallest wrinkle appearance coefficient as a reference frame, filter the wrinkle area in the reference frame, and obtain wrinkle appearance elimination parameters based on the difference in wrinkle appearance coefficients of the key frames before and after filtering; for each facial image, obtain the smoothing processing coefficient of each facial image based on the wrinkle appearance elimination parameters and the wrinkle appearance associated parameters and perform smoothing processing to obtain the facial age reduction and reshaping result.

[0011] Furthermore, the method for acquiring the face image includes:

[0012] Video frames containing faces of film and television characters are obtained, the number of optical flow movements of each video frame is obtained, and video frames whose number of optical flow movements is less than a preset threshold are used as the face images.

[0013] Furthermore, the method for obtaining the eye region includes:

[0014] The eye area is obtained based on Haar-like features using a Haar classifier, and the polar coordinates of the edge contour points of the eye area are obtained. The intersection of the straight lines where the polar coordinates are located is taken as the center point.

[0015] Furthermore, the method for obtaining the organ expression change parameter includes:

[0016] The difference between the average value of the distances between adjacent edge contour points and the positive integer 1 is used as the discontinuity;

[0017] The length of the shortest reference line is normalized and then multiplied by the discontinuity to obtain the organ expression change parameter.

[0018] Furthermore, the method for obtaining the wrinkle appearance coefficient includes:

[0019] The wrinkle appearance coefficient is obtained by normalizing the ratio of the area proportion of the associated wrinkle line in the wrinkle area to the average pixel value of the wrinkle area.

[0020] Furthermore, the method for obtaining the overall wrinkle appearance coefficient includes:

[0021] The average value of the wrinkle appearance coefficients of all wrinkle areas in the face image is used as the overall wrinkle appearance coefficient.

[0022] Furthermore, the method for obtaining wrinkle appearance-related parameters includes:

[0023] normalizing the ratio of the organ expression change parameter difference and the overall wrinkle appearance coefficient difference to obtain the wrinkle appearance correlation parameter.

[0024] Further, the method for obtaining the wrinkle appearance elimination parameter comprises:

[0025] normalizing the difference between the wrinkle appearance coefficient of the key frame and the wrinkle appearance coefficient of the filtered image of the key frame to obtain the wrinkle appearance elimination parameter.

[0026] Further, the method for obtaining the smoothing processing coefficient comprises:

[0027] taking the sum of the normalized wrinkle appearance correlation parameter and the positive integer 1 as the adjustment multiple, and taking the product of the adjustment multiple and the wrinkle appearance elimination parameter as the smoothing processing coefficient.

[0028] The application further provides a film and television role face rejuvenation system based on deep learning, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the steps of any one of the film and television role face rejuvenation methods based on deep learning.

[0029] The application has the following beneficial effects:

[0030] The application integrates a face image processing algorithm in a deep convolutional network, and obtains various image features in the face image through the deep convolutional network. In consideration of the fact that the eyes are the main change organs when wrinkles are generated due to facial expressions, the eye region is identified, and the organ expression change parameter is determined in combination with the shape of the eye region contour. The application aims to appropriately smooth the wrinkles based on the facial expressions, to avoid over-smoothing and under-smoothing, and thus the correlation between the eye region and the wrinkles needs to be determined, and thus the associated wrinkle lines in each wrinkle region are determined through the positional relationship between the organs and the wrinkle regions in the wrinkle region, and the wrinkle appearance coefficient of each wrinkle region is obtained, that is, the greater the wrinkle appearance coefficient, the more obvious the wrinkles in the face image, and the greater the degree of smoothing. Further considering the information change between adjacent frame face images, the wrinkle appearance correlation parameter of each frame face image is effectively determined. The wrinkle appearance elimination parameter obtained by taking the face image with the minimum wrinkle appearance coefficient as the reference frame can be used as the basis of the smoothing parameter, because it is the information obtained by taking the reference frame with the minimum wrinkle appearance coefficient, and thus other face images can be enlarged on the basis of the reference frame in combination with the wrinkle appearance correlation parameter, so as to obtain the smoothing processing parameter of each face image and effectively rejuvenate the face. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 A flowchart of a method for processing facial images using a deep convolutional network in a method for reducing the age of film and television character faces based on deep learning provided by one embodiment of the present invention;

[0033] Figure 2 A schematic diagram of wrinkle area identification provided by one embodiment of the present invention;

[0034] Figure 3 A schematic diagram of an eye region outline provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] To further illustrate the technical means and effectiveness of the present invention to achieve the intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail a method and system for reshaping the face of a film or television character based on deep learning, including its specific implementation, structure, features, and effectiveness. In the following description, different references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.

[0036] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0037] The present invention aims to process facial images using a deep convolutional network, extracting features from the facial image and performing operations such as convolution and pooling, ultimately determining the facial de-aging and reshaping results. The deep neural network used in the present invention is the AlexNet network structure, which is a well-known technical means for those skilled in the art and will not be described in detail here. The following briefly describes the characteristics of the network structure and the data processing method:

[0038] The AlexNet network structure consists of 5 convolutional layers and 3 fully connected layers. Each convolutional layer contains a convolution kernel, a bias term, a ReLU activation function, and a local response normalization (LRN) module. The 1st, 2nd, and 5th convolutional layers are followed by a maximum pooling layer.

[0039] After the face image of a movie character is input into the network model, each convolution layer is based on the image and uses a filter to slide once in the width or height dimension of the image, at this time the filter and the input image form a corresponding relationship in space, and an inner product operation is performed at the corresponding position in the space to obtain a pixel value in the output feature map, and then multiple slides are performed until a complete feature map is obtained, when all filters complete feature extraction of the image, the feature maps are stacked in the depth dimension to form the output of the convolution layer. The wrinkle appearance coefficient and the organ expression change parameter obtained by the embodiment of the present application are weights during image convolution processing, and the space dimension of the feature map is reduced in the pooling layer through the weights, so that the feature information can be retained. After feature extraction is performed on each face image, adjacent frame face images can be analyzed to determine the wrinkle appearance correlation parameter, and then the wrinkle appearance elimination parameter is used to determine the smoothing processing coefficient of the final face image, and smoothing processing is performed to output the face rejuvenation result.

[0040] Before the face image is input into the deep convolution network, a series of preprocessing processes should be performed on the face image in the embodiment of the present application to facilitate feature extraction in the network. First, the face image of a movie character should be determined in a movie video, and then the wrinkle area and the eye area in the face image are identified. The wrinkle area is the area that needs to be smoothed in the embodiment of the present application, because the main wrinkles produced on the face, such as forehead wrinkles and crow's feet, include static wrinkles and dynamic wrinkles, and the features of static wrinkles are relatively fixed, while dynamic wrinkles are affected by facial expressions. The eyes are important facial organs, and the facial expressions that produce wrinkles are often accompanied by changes in the eyes, for example, when a person smiles, the eyes will be narrowed, and then obvious wrinkles will be produced at the corners of the eyes. Therefore, the eye area is identified in the embodiment of the present application to facilitate subsequent processing of the face image in the deep convolution network.

[0041] Preferably, in an embodiment of the present application, considering that the facial expressions of a movie character in a movie video include useless image frames when continuous frames are considered, such image frames change more and faster, and because the facial features are weak, the face rejuvenation of the face is not needed, therefore, based on motion analysis, the embodiment of the present application obtains video frames containing the face of a movie character, obtains the number of optical flow movements of each video frame, and regards the video frames with the number of optical flow movements less than a preset threshold as the face image. It should be noted that the threshold of the number of optical flow movements can be set according to the specific content of the movie video, and in the embodiment of the present application, the number of optical flow movements of all video frames containing the face of a movie character is arranged from small to large, and the number of optical flow movements at the top thirty percent position is selected as the threshold.

[0042] In the embodiment of the present invention, the wrinkle region extraction method can be based on the content of "Research on Wrinkle Detection and Quantitative Evaluation of Facial Images", combining the filtering results of Gabor filter bank and Frangi filter as image features, and extracting features of wrinkles of different coarseness and fineness. Figure 2 , which shows a schematic diagram of wrinkle area identification provided by an embodiment of the present invention.

[0043] Preferably, in an embodiment of the present invention, the face image is regarded as a whole two-dimensional pixel matrix, and a face pattern space is constructed from a statistical point of view through a large number of face image samples, and then the Haar classifier is used to obtain the eye area based on Haar-like features, and the polar coordinate points of the edge contour points of the eye area are obtained, and the intersection of the straight lines where the polar coordinate points are located is used as the center point. It should be noted that the Haar classifier will eventually obtain multiple strange faces, because the eye area has unique characteristics compared to other organs. For example, compared with the bridge of the nose, the pixel points of the eye part are smaller, and the pixel values ​​of the bridge of the nose part are higher. In addition, there are different pixel value distributions of the pupil, white of the eye, and upper and lower eyelids inside the eye. The eye area can be extracted separately through contour detection and threshold segmentation technology. The specific means are technical means well known to those skilled in the art and will not be described here. Please refer to Figure 3 , which shows a schematic diagram of an eye region outline provided by an embodiment of the present invention, Figure 3 The origin O in the upper left corner is the origin of the image coordinate system. The two coordinate axes of the image coordinate system are i and j. By counting the polar coordinates of all edge contour points, two vertical extreme lines can be obtained. The center pixel point corresponding to the extreme line is recorded as the center point.

[0044] See also Figure 1 , which shows a flowchart of a method for processing facial images using a deep convolutional network in a method for reducing the age of film and television character faces based on deep learning provided by one embodiment of the present invention. The method includes:

[0045] Step S1: taking the line between the center point and each edge contour point as a reference line; and obtaining the organ expression change parameter of the eye region according to the discontinuity of the edge contour points and the length of the reference line.

[0046] The eye area is an area with obvious facial features, and its contour features have a clear correlation with the appearance of wrinkles. For example, when a character smiles, his eyes will squint, and there will be a slight upward change in the corners of the eyes. At this time, the distance between the top and bottom of the eyes and the center point is small, and obvious wrinkles will appear at the corners of the eyes; for example, when a character raises his eyebrows, his eyes will open wide, and the distance between the top and bottom of the eyes and the center point is small, and the eye contour changes significantly, and forehead wrinkles will appear on the character's forehead. Therefore, the embodiment of the present invention connects the center point with each edge contour point to obtain a reference line, that is, an eye area contains multiple reference lines. Further considering the discontinuity of the eye area, the greater the discontinuity, the greater the degree of deformation of the eye contour, that is, the greater the degree of expression change at this time. Therefore, the organ expression change parameter of the eye area can be obtained based on the discontinuity of the edge contour point and the length of the reference line. The organ expression change parameter characterizes the expression change characteristics of the eye area in the current face image.

[0047] Preferably, in one embodiment of the present invention, the method for acquiring organ expression change parameters includes:

[0048] The difference between the average distance between adjacent edge contour points and the positive integer 1 is used as the discontinuity. In other words, the present invention assumes that, ideally, the distance between adjacent edge contour points should be 1, resulting in a calm expression and no noticeable change in the eye contour. Therefore, the greater the difference between the average distance between adjacent edge contour points and the positive integer 1, the more pronounced the eye contour changes, and the more pronounced the eye's organ expression change parameter should be.

[0049] The length of the shortest reference line is normalized and then multiplied by the discontinuity to obtain the organ expression change parameter. In the embodiment of the present invention, the length of the shortest reference line is used as the expression feature of the eye at that time. The longer the length of the shortest reference line is, the greater the degree of opening and closing of the eyes, and the greater the organ expression change parameter. It should be noted that the method for normalizing the length of the shortest reference line in the embodiment of the present invention is: using the length of the shortest reference line as the numerator and the length of the average reference line as the denominator to obtain the normalized result.

[0050] In one embodiment of the present invention, in order to facilitate subsequent data processing, normalization processing needs to be performed after obtaining the organ expression change parameters, and a normalization result with a value range between 0 and 1 is obtained using a softsign function.

[0051] Step S2: Extend the reference line and use the intersection with the nearest wrinkle area as the associated wrinkle point; in each wrinkle area, use the line connecting the associated wrinkle points as the associated wrinkle line; and obtain the wrinkle appearance coefficient of each wrinkle area based on the proportion of the associated wrinkle line in the wrinkle area and the pixel value of the wrinkle area.

[0052] Wrinkles on a character's face can be classified into two types: static and dynamic. Dynamic wrinkles arise from changes in facial expression. Therefore, to eliminate dynamic wrinkles using appropriate smoothing parameters, it's necessary to determine the correlation between the eye and wrinkle regions. Therefore, based on the relationship between the eye and wrinkle locations, this embodiment of the present invention extends a reference line and uses its intersection with the nearest wrinkle region as the associated wrinkle point. The line connecting the associated wrinkle points is then called the associated wrinkle line.

[0053] The greater the proportion of associated wrinkle lines in the wrinkle area, the more obvious the dynamic wrinkles are. At the same time, the darker the pixel value in the wrinkle area, the more obvious the wrinkles are relative to normal skin. Therefore, the wrinkle appearance coefficient of each wrinkle area can be obtained based on the proportion of associated wrinkle lines in the wrinkle area and the pixel value of the wrinkle area.

[0054] Preferably, in an embodiment of the present invention, the method for obtaining the wrinkle appearance coefficient includes:

[0055] The wrinkle appearance coefficient is obtained by normalizing the ratio of the area ratio of the associated wrinkle line in the wrinkle region to the average pixel value of the wrinkle region. It should be noted that the area ratio in this embodiment of the present invention is the ratio of the number of pixels on the associated wrinkle line to the number of pixels within the minimum bounding rectangle of the wrinkle region.

[0056] Step S3: Obtain the overall wrinkle appearance coefficient of the current face image based on the wrinkle appearance coefficients of all wrinkle areas, and obtain the wrinkle appearance correlation parameters of each face frame based on the differences in the overall wrinkle appearance coefficients between adjacent frames of face images and the differences in organ expression change parameters.

[0057] After processing in step S1, each wrinkle region in the facial image corresponds to a wrinkle appearance coefficient. Therefore, all wrinkle appearance coefficients can be statistically calculated to obtain the overall wrinkle appearance coefficient for the current facial image. The overall wrinkle appearance coefficient reflects the degree of wrinkle appearance in the current facial image, particularly the degree of wrinkle appearance caused by facial expression. Because facial expressions are dynamic, further analysis can be performed to examine the differences in overall wrinkle appearance coefficients between facial images at adjacent moments, as well as differences in organ expression variation parameters. This information reveals the correlation between wrinkle appearance coefficients after facial expression changes between adjacent frames, thereby obtaining the wrinkle appearance correlation parameter for each facial image frame. A larger wrinkle appearance correlation parameter indicates that the wrinkle region in that facial image frame is likely caused by drastic or frequent changes in facial expression, exhibiting more pronounced wrinkle characteristics and requiring a larger smoothing parameter.

[0058] Preferably, in one embodiment of the present invention, the method for obtaining wrinkle appearance-related parameters includes:

[0059] The ratio of the difference in organ expression change parameters to the difference in the overall wrinkle appearance coefficient is normalized to obtain the wrinkle appearance correlation parameter. In this embodiment of the present invention, for each facial image frame, the difference between the organ expression change parameters of the previous frame and the next frame is used as the organ expression change parameter difference; the overall wrinkle appearance coefficient difference is obtained in a similar manner. Specifically, the greater the difference in facial expression change between the previous and next frames, while the difference in the overall wrinkle appearance coefficient is smaller, indicating a greater correlation between the facial expression change and the wrinkle state, and the need to further emphasize the weighting features in the dynamic wrinkle smoothing process.

[0060] Step S4: Among multiple frames of facial images, select the smallest wrinkle appearance coefficient as the reference frame, filter the wrinkle area in the reference frame, and obtain the wrinkle appearance elimination parameters based on the difference in wrinkle appearance coefficients of the key frames before and after filtering; for each facial image, obtain the smoothing processing coefficient of each facial image based on the wrinkle appearance elimination parameters and the wrinkle appearance related parameters and perform smoothing processing to obtain the facial age reduction and reshaping result.

[0061] When reshaping the faces of film and television characters to reduce their age, over-smoothing of wrinkles should be avoided as much as possible. Over-smoothing wrinkles will cause the face to lose its natural feel, so some dynamic lines should be retained as much as possible to avoid over-smoothing. Therefore, it is necessary to determine a suitable smoothing parameter as basic data. Adjusting the smoothing parameter based on this basic data can prevent over-smoothing while also achieving an adaptive effect on wrinkle smoothing. Because the wrinkle appearance coefficient can reflect the degree of wrinkle visibility and the degree of dynamic line appearance, the embodiment of the present invention selects the smallest wrinkle appearance coefficient as the reference frame in multiple frames of facial images, filters the wrinkle area in the reference frame, and obtains the wrinkle appearance elimination parameter based on the difference in the wrinkle appearance coefficients of the key frames before and after filtering. That is, the wrinkle appearance elimination parameter is used as the basic smoothing parameter, and the smoothing parameters of other facial images can be adjusted and changed based on this to achieve adaptive smoothing while avoiding over-smoothing.

[0062] The embodiment of the present invention further uses wrinkle appearance related parameters as weights, and adjusts the wrinkle appearance elimination parameters to obtain a smoothing processing coefficient for each facial image. After smoothing the facial image using the smoothing processing coefficient, a facial age reduction and reshaping result can be obtained.

[0063] Preferably, in an embodiment of the present invention, the method for obtaining wrinkle appearance and elimination parameters includes:

[0064] The wrinkle appearance coefficient of the key frame and the wrinkle appearance coefficient of the filtered image of the key frame are normalized to obtain a wrinkle appearance elimination parameter. In this embodiment of the present invention, the ratio of the difference to the wrinkle appearance coefficient of the key frame is used as the final normalization result.

[0065] Preferably, in an embodiment of the present invention, the method for obtaining the smoothing coefficient includes:

[0066] The adjustment factor is the sum of the normalized wrinkle appearance parameter and the positive integer 1. That is, the adjustment factor has a value range of 1 to 2. The product of the adjustment factor and the wrinkle appearance elimination parameter is used as the smoothing coefficient.

[0067] It should be noted that the method for filtering and smoothing wrinkles in the embodiment of the present invention may adopt an existing filtering method, such as a non-mean filtering algorithm, and the specific method is not limited thereto.

[0068] In summary, the embodiment of the present invention identifies wrinkle areas and eye areas in the facial images of film and television characters, inputs the facial images into a deep convolutional network, and the deep convolutional network performs feature extraction of the organ expression change parameters of the eye area, as well as feature extraction of the wrinkle appearance coefficient of the facial image. By analyzing adjacent frames, the wrinkle appearance associated parameters of the facial image are determined. The minimum wrinkle appearance coefficient is used as a reference frame to obtain the wrinkle appearance elimination coefficient, and the wrinkle appearance elimination parameters are adjusted using the wrinkle appearance associated parameters to achieve adaptive smoothing of the facial image and obtain a facial age reduction and reshaping result. The present invention avoids the problems of overfitting and unnatural age reduction and reshaping results by adjusting the smoothing parameters of the character's facial image, thereby improving the quality of the age reduction and reshaping results.

[0069] Based on the same inventive concept, the present invention also proposes a deep learning-based facial de-aging and reshaping system for film and television characters, comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the computer program, it implements any one of the steps of a deep learning-based facial de-aging and reshaping method for film and television characters.

[0070] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0071] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A method for reducing the age of film and television character faces based on deep learning, characterized in that: The method comprises: Obtain multiple frames of facial images of film and television characters, identify wrinkle areas and eye areas in the facial images, and identify the edge contours and center points of the eye areas; input the facial images into a deep convolutional network to obtain a facial age reduction and reshaping result; the method for processing facial images using the deep convolutional network includes: Step S1: using the line between the center point and each edge contour point as a reference line; obtaining the organ expression change parameter of the eye region according to the discontinuity of the edge contour points and the length of the reference line; Step S2: Extending the reference line and defining its intersection with the nearest wrinkle region as an associated wrinkle point; defining a line connecting the associated wrinkle points as an associated wrinkle line in each wrinkle region; and obtaining a wrinkle appearance coefficient for each wrinkle region based on the proportion of the associated wrinkle line in the wrinkle region and the pixel value of the wrinkle region. Step S3: Obtaining the overall wrinkle appearance coefficient of the current facial image based on the wrinkle appearance coefficients of all wrinkle areas, and obtaining the wrinkle appearance correlation parameters of each facial image frame based on the differences in the overall wrinkle appearance coefficients and the differences in organ expression change parameters between adjacent facial images; Step S4: Selecting the smallest wrinkle appearance coefficient from the multiple frames of facial images as a reference frame, filtering the wrinkle area in the reference frame, and obtaining wrinkle appearance elimination parameters based on the difference in wrinkle appearance coefficients between the key frames before and after filtering; obtaining a smoothing coefficient for each facial image based on the wrinkle appearance elimination parameters and the wrinkle appearance association parameters, and performing smoothing to obtain a facial age reduction and reshaping result; The method for obtaining the wrinkle appearance coefficient includes: Normalizing the ratio of the area ratio of the associated wrinkle line in the wrinkle area to the average pixel value of the wrinkle area to obtain the wrinkle appearance coefficient; The method for obtaining the overall wrinkle appearance coefficient includes: taking the average value of the wrinkle appearance coefficients of all wrinkle areas in the face image as the overall wrinkle appearance coefficient; The method for obtaining wrinkle appearance related parameters includes: Normalizing the ratio of the organ expression change parameter difference to the overall wrinkle appearance coefficient difference to obtain the wrinkle appearance correlation parameter; The method for obtaining wrinkle appearance and elimination parameters includes: Normalizing the difference between the wrinkle appearance coefficient of the key frame and the wrinkle appearance coefficient of the filtered image of the key frame to obtain the wrinkle appearance elimination parameter; The method for obtaining the smoothing coefficient includes: The sum of the normalized wrinkle appearance associated parameter and the positive integer 1 is used as the adjustment magnification, and the product of the adjustment magnification and the wrinkle appearance elimination parameter is used as the smoothing coefficient.

2. The method for reducing the age of a film and television character's face based on deep learning according to claim 1, characterized in that: The method for acquiring the face image includes: Video frames containing faces of film and television characters are obtained, the number of optical flow movements of each video frame is obtained, and video frames whose number of optical flow movements is less than a preset threshold are used as the face images.

3. The method for reducing the age of a film and television character's face based on deep learning according to claim 1, characterized in that: The method for obtaining the eye region includes: The eye area is obtained based on Haar-like features using a Haar classifier, and the polar coordinates of the edge contour points of the eye area are obtained. The intersection of the straight lines where the polar coordinates are located is taken as the center point.

4. The method for reducing the age of a film and television character's face based on deep learning according to claim 1, characterized in that: The method for obtaining the organ expression change parameter includes: The difference between the average value of the distances between adjacent edge contour points and the positive integer 1 is used as the discontinuity; The length of the shortest reference line is normalized and then multiplied by the discontinuity to obtain the organ expression change parameter.

5. A deep learning-based system for reshaping the face of a film or television character to reduce age, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of a method for reducing the age of a face of a film and television character based on deep learning as described in any one of claims 1 to 4 are implemented.

Images