WO2019041660A1 - 人脸去模糊方法及装置 - Google Patents
人脸去模糊方法及装置 Download PDFInfo
- Publication number
- WO2019041660A1 WO2019041660A1 PCT/CN2017/117166 CN2017117166W WO2019041660A1 WO 2019041660 A1 WO2019041660 A1 WO 2019041660A1 CN 2017117166 W CN2017117166 W CN 2017117166W WO 2019041660 A1 WO2019041660 A1 WO 2019041660A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- face
- image
- dimensional
- processed
- grid
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 91
- 238000012545 processing Methods 0.000 claims description 23
- 230000015654 memory Effects 0.000 claims description 18
- 238000010586 diagram Methods 0.000 claims description 9
- 230000036544 posture Effects 0.000 abstract description 20
- 230000000694 effects Effects 0.000 abstract description 7
- 210000001508 eye Anatomy 0.000 description 4
- 210000004709 eyebrow Anatomy 0.000 description 4
- 210000000214 mouth Anatomy 0.000 description 4
- 210000001331 nose Anatomy 0.000 description 4
- 239000013598 vector Substances 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000004590 computer program Methods 0.000 description 3
- 210000000056 organ Anatomy 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 210000000887 face Anatomy 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 230000014616 translation Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011840 criminal investigation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2200/00—Indexing scheme for image data processing or generation, in general
- G06T2200/04—Indexing scheme for image data processing or generation, in general involving 3D image data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
- G06T2207/30201—Face
Definitions
- the present invention relates to the field of image processing technologies, and in particular, to a face deblurring method and apparatus.
- a fuzzy region is first extracted from a face image to be processed, and then the fuzzy region is processed by a correlation algorithm to recover an implicit clear image from the blurred image.
- the embodiment of the present invention provides a method and a device for deblurring a face to solve the problem of poor deblurring processing of a face image in the prior art.
- a first aspect of the present invention provides a method for deblurring a face, comprising the steps of:
- the first grid dictionary is obtained by aligning and dividing the first two-dimensional library according to the face template, and the first two-dimensional library is a blurred image created by the first three-dimensional face library obtained by the three-dimensional reconstruction method. 2D face gallery;
- the second mesh dictionary is a second two-dimensional library according to the After the face template is aligned and divided, the second two-dimensional library is a two-dimensional face library of clear images created by the second three-dimensional face library obtained by the three-dimensional reconstruction method, and the blurred image and the clear image are one by one. correspond;
- each mesh of the divided face image to be processed is matched with a mesh of the first mesh dictionary to obtain multiple fuzzy corresponding to each mesh of the to-be-processed face image.
- the grid includes the following steps:
- querying, in the second grid dictionary, a plurality of clear grids corresponding to the plurality of fuzzy grids one by one including the following steps:
- a clear grid corresponding to the blurred mesh is queried in the second grid dictionary according to the coordinates.
- the defuzzifying the mesh of the to-be-processed face image according to the clear grid includes the following steps:
- first three-dimensional face library and a second three-dimensional face library obtained by using a three-dimensional reconstruction method, wherein the first three-dimensional face gallery and the second three-dimensional face gallery respectively are a plurality of blurred images and corresponding clear images Dimensional expansion diagram;
- the attitude parameter is an angle ( ⁇ x , ⁇ y , ⁇ z ) of the to-be-processed face image in a three-dimensional space;
- ⁇ x is an offset angle of the to-be-processed face image in the x direction
- ⁇ y is an offset angle of the to-be-processed face image in the y direction
- ⁇ z is the to-be-processed face image The angle of the offset in the z direction.
- a second aspect of the present invention provides a face deblurring apparatus, including:
- a first acquiring unit configured to acquire a to-be-processed face image
- a dividing unit configured to align the image of the face to be processed onto the face template, and mesh the same;
- a matching unit configured to match each of the meshes of the divided face images to be matched with the mesh of the first grid dictionary, to obtain multiple corresponding to each mesh of the to-be-processed face image a fuzzy grid, wherein the first grid dictionary is obtained by aligning and dividing the first two-dimensional library according to the face template, and the first two-dimensional library is established by using a three-dimensional reconstruction method obtained by using a three-dimensional reconstruction method.
- a query unit configured to query, according to the fuzzy mesh, a plurality of clear meshes corresponding to the plurality of fuzzy meshes in a second grid dictionary, wherein the second mesh dictionary is a second
- the two-dimensional library is obtained by aligning and dividing the face template, and the second two-dimensional library is a two-dimensional face library of clear images established by the three-dimensional face library obtained by the three-dimensional reconstruction method, the blurred image and the clear One-to-one correspondence of images;
- a processing unit configured to generate a clear image of the to-be-processed face image according to the clear grid that is queried.
- the matching unit includes:
- a second acquiring unit configured to respectively acquire each grid of the to-be-processed face image and pixels of each grid of the first grid dictionary
- a calculating unit configured to calculate, according to the acquired pixels, an Euclidean distance of a pixel similarity between each mesh of the to-be-processed face image and each mesh of the first grid dictionary
- a third acquiring unit configured to acquire M fuzzy meshes that match each mesh of the to-be-processed face image according to the calculated Euclidean distance.
- a third aspect of the present invention provides an image processing apparatus including at least one processor; and a memory communicably coupled to the at least one processor; wherein the memory stores instructions executable by the one processor, The instructions are executed by the at least one processor to cause the at least one processor to perform the face deblurring method of any of the first aspects of the invention.
- a fourth aspect of the invention provides a non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the first aspect or the first aspect The face deblurring method described in any of the alternatives.
- a fifth aspect of the invention provides a computer program product comprising a computing program stored on a non-transitory computer readable storage medium, the computer program comprising program instructions, when the program instructions are executed by a computer
- the computer is caused to perform the first aspect or the face deblurring described in any of the alternative aspects of the first aspect.
- a method for blurring a face includes the following steps: acquiring a face image to be processed; aligning the image of the face to be processed onto a face template, and meshing the face image; Each mesh of the to-be-processed face image is matched with a mesh of the first mesh dictionary to obtain a plurality of fuzzy meshes corresponding to each mesh of the to-be-processed face image, wherein the The first grid dictionary is obtained by aligning and dividing the first two-dimensional library according to the face template, and the first two-dimensional library is a two-dimensional person of the blurred image created by the first three-dimensional face library obtained by the three-dimensional reconstruction method.
- the second two-dimensional library is a two-dimensional face library of clear images created by the second three-dimensional face library obtained by the three-dimensional reconstruction method, the blurred image and the clear Image one-to-one correspondence
- the query to the mesh clear, sharp image to be processed to generate the face image can process face images of different postures, and has a better face deblurring effect.
- the method for deblurring a face according to the embodiment of the present invention wherein defuzzifying the mesh of the image to be processed according to the clear mesh includes the following steps: acquiring the clear mesh Pixels of the to-be-processed face image are processed such that pixels of each grid of the to-be-processed face image are sums of pixels of the plurality of clear grids.
- the grid pixel obtained by weighting the same position on the face of different clear faces is used to replace the original fuzzy mesh. Pixels, with better face deblurring effect.
- the method for deblurring a face includes the following steps: acquiring a first three-dimensional face gallery and a second three-dimensional face gallery obtained by using the three-dimensional reconstruction method,
- the first three-dimensional face library and the second three-dimensional face gallery are respectively a plurality of blurred images and a corresponding two-dimensional columnar expansion image; the posture parameters of the to-be-processed face image are configured;
- a parameter respectively establishing a corresponding first two-dimensional library and a second two-dimensional library in the first three-dimensional face gallery and the second three-dimensional face gallery.
- the face deblurring method provided by the embodiment of the invention can set the posture parameter of the face image to be processed in the space according to the user, so that the two-dimensional corresponding parameter parameters can be obtained from the histogram dictionary.
- the image dictionary in turn, can handle the face deblurring process with gestures in the video surveillance scene.
- the face deblurring apparatus includes: a first acquiring unit, configured to acquire a to-be-processed face image; and a dividing unit, configured to align the to-be-processed face image onto the face template, and Performing meshing; a matching unit, configured to match each of the divided meshes of the image to be processed with a mesh of the first mesh dictionary to obtain each network of the image of the face to be processed a plurality of fuzzy meshes corresponding to the cells, wherein the first mesh dictionary is obtained by aligning and dividing the first two-dimensional library according to the face template, and the first two-dimensional library is a three-dimensional image obtained by using a three-dimensional reconstruction method.
- the face deblurring device provided by the embodiment of the invention can process face images of different postures and has a better face deblurring effect.
- FIG. 1 is a schematic diagram showing a frontal face turning to another gesture face in three-dimensional face reconstruction
- Figure 2 shows a three-dimensional shape model of a human face
- Figure 3 shows a two-dimensional columnar development of a three-dimensional face texture
- FIG. 4 is a flow chart showing a specific schematic of a method for deblurring a face in Embodiment 1 of the present invention.
- FIG. 5 is a flow chart showing a specific schematic of a method for deblurring a face in Embodiment 2 of the present invention.
- FIG. 6 is a flow chart showing a specific schematic of a method for deblurring a face in Embodiment 3 of the present invention.
- Figure 7 is a block diagram showing a specific schematic of a face deblurring apparatus in Embodiment 4 of the present invention.
- FIG. 8 is a block diagram showing still another detailed schematic diagram of a face deblurring apparatus in Embodiment 4 of the present invention.
- Fig. 9 is a block diagram showing a concrete schematic diagram of an image processing apparatus in Embodiment 5 of the present invention.
- the process of three-dimensional face reconstruction is to convert the frontal face image into a face image of any other angle by using the three-dimensional face model, and the specific process is as shown in FIG. 1 , that is, for a frontal face image, Given the pose parameters, the face image in the corresponding pose can be obtained.
- the three-dimensional face model contains the following four aspects of information:
- a frontal human face three-dimensional shape S (x 1 , y 1 , z 1 , ..., x n , y n , z n ) T , where n represents the number of shape vertices;
- each shape point of the two-dimensional shape can correspond to the texture pixel value, as shown in FIG. 2, the three-dimensional shape model of the human face;
- Step 1 Use ASM to locate 68 face key points
- Step 2 Use the 68 key points to align the current human face two-dimensional shape to the three-dimensional shape in the model. Because it is a frontal face, the rotation angle in the Z direction is 0, so the parameters ( ⁇ x , ⁇ y , ⁇ z , are required. ⁇ x, ⁇ y, ⁇ z, s), where ⁇ x , ⁇ y , ⁇ z represent the rotation angles of x, y, and z, respectively, ⁇ x, ⁇ y, and ⁇ z represent translations in three directions of x, y, and z, respectively. And s represent a scaling factor;
- Step 3 Reusing the rotation, translation and scaling parameters calculated in step 2 to rotate, translate and scale each vertex of the shape S described above, that is, the current shape is aligned to the standard shape in the model;
- Step 4 Finally, using the information in step (2) and step (3), combined with Kriging interpolation to obtain a two-dimensional columnar expansion map of the complete three-dimensional face texture map as shown in FIG. 3;
- Step 5 When the user inputs an angle parameter ( ⁇ x , ⁇ y , ⁇ z ) that rotates the current front face, the aligned shape vertices are rotated (the inverse process of step 2);
- Step 6 According to the parameter calculated in step 5, combined with the two-dimensional columnar expansion map of the three-dimensional face texture map generated in step 4, the target image is obtained by Kriging interpolation.
- the two-dimensional texture library under the posture can be transformed, and the different face processing models are trained online by using the two-dimensional texture library to satisfy different image processing of any posture.
- the present embodiment provides a face deblurring method for use in a face deblurring device. As shown in FIG. 4, the face deblurring method includes the following steps:
- Step S11 Acquire a face image to be processed.
- the face image in this embodiment may be previously stored in the face deblurring device, or the face deblurring device may acquire an image obtained from the outside world in real time, or may be an image extracted from a video by the face deblurring device.
- step S12 the image of the face to be processed is aligned onto the face template and meshed.
- the embodiment of the present invention divides the face image into small grids, and each organ is composed of a plurality of meshes.
- the face template in this embodiment is a reference for dividing the image of the face to be processed.
- the face image to be processed is firstly normalized and dimensioned, that is, the face, the eyebrow, the nose, the mouth, and the corresponding organ of the face template are in the face. The same position, thus ensuring the standardization and accuracy of the division.
- the image of the face to be processed may be divided into meshes of m rows and n columns, where m is the number of meshes in the vertical direction, and n is the number of meshes in the horizontal direction.
- the specific values of m and n can be set according to the actual precision of the processed face image to be processed, and may or may not be equal.
- Step S13 matching each mesh of the divided face image to be matched with the mesh of the first mesh dictionary, and obtaining a plurality of fuzzy meshes corresponding to each mesh of the to-be-processed face image, where
- the first grid dictionary is obtained by aligning the first two-dimensional library according to the face template, and the first two-dimensional library is a two-dimensional face gallery of the blurred image created by the first three-dimensional face gallery obtained by the three-dimensional reconstruction method.
- the first three-dimensional face gallery is a face dictionary library of the blurred image of the S face, that is, the two-dimensional face reconstruction method of the corresponding blurred image is obtained by using the three-dimensional face reconstruction method described above;
- the first two-dimensional library is extracted from the two-dimensional column expanded image with respect to the deflection angle of the face image with respect to the front image.
- the first grid dictionary is obtained by aligning the first two-dimensional library according to the face template alignment; wherein the blurred image of the S face is aligned and dimensioned, that is, each face is aligned.
- the eyes, the eyebrows, the nose, and the mouth are substantially at the same position in the picture, so that the coordinates of the mesh in the first grid dictionary and the corresponding position of the image to be processed are the same on the face template.
- Step S14 according to the fuzzy mesh, querying, in the second grid dictionary, a plurality of clear meshes corresponding to the plurality of fuzzy meshes, wherein the second mesh dictionary is a second two-dimensional library according to the face template After the alignment is divided, the second two-dimensional library is a two-dimensional face gallery of clear images created by the second three-dimensional face gallery obtained by the three-dimensional reconstruction method, and the blurred images are in one-to-one correspondence with the clear images.
- the second three-dimensional face database is a face dictionary library of a clear image of the S face, that is, a two-dimensional column expansion view of the corresponding clear image is obtained by using the three-dimensional face reconstruction method described above;
- a second two-dimensional library is extracted from the two-dimensional column expanded image with respect to a deflection angle of the face image with respect to the front image.
- the second grid dictionary is obtained by aligning the second two-dimensional library according to the face template alignment, that is, the first grid dictionary, the second grid dictionary, and the grid corresponding to the position of the to-be-processed face image are on the face template.
- the coordinates are the same.
- the blurred image and the sharp image described in the embodiments of the present invention and the fuzzy mesh and the clear mesh mentioned later are opposite, for example, the clear image represents an image that can be quickly recognized by the human eye, and Specifically, it can be defined by some parameters of the image (for example, pixel values), and the blurred image is similar.
- Step S15 Generate a clear image of the image of the face to be processed according to the clear grid that is queried.
- the grid in the face image to be processed may be directly replaced by the clear grid that is queried, or the grid pixel in the image of the face to be processed may be replaced by the pixel of the clarified grid.
- the present embodiment provides a face deblurring method for use in a face deblurring device. As shown in FIG. 5, the face deblurring method includes the following steps:
- Step S21 Acquire a face image to be processed. It is the same as step S11 in Embodiment 1, and will not be described again.
- step S22 the image of the face to be processed is aligned onto the face template and meshed. It is the same as step S12 in Embodiment 1, and will not be described again.
- Step S23 matching each mesh of the divided face image to be matched with the mesh of the first mesh dictionary, and obtaining a plurality of fuzzy meshes corresponding to each mesh of the to-be-processed face image, where
- the first grid dictionary is obtained by aligning the first two-dimensional library according to the face template, and the first two-dimensional library is a two-dimensional face gallery of the blurred image created by the first three-dimensional face gallery obtained by the three-dimensional reconstruction method.
- the Euclidean distance between the grid of the image of the face to be processed and the pixel of the grid of the first grid dictionary is calculated, and the first matching with the grid of the image of the face to be processed is found according to the Euclidean distance.
- the grid in the grid dictionary is calculated.
- step S23 specifically includes the following steps:
- Step S231 respectively acquiring each grid of the face image to be processed and the pixels of each grid of the first grid dictionary.
- the pixels of the grid of the face image to be processed and the pixels of the grid of the first grid dictionary are obtained by summing the pixel values of all the pixels in each grid and averaging; Since the image of the face to be processed and the first grid dictionary are aligned to the face template for meshing, the number of meshes to be processed by the face image to be processed and the first grid dictionary is desired.
- the pixels of the grid of the face image to be processed and the pixels of the grid of the first grid dictionary are represented by pixel value vectors composed of all the pixels in the grid.
- Step S232 calculating, according to the acquired pixels, the Euclidean distance of each pixel of the to-be-processed face image and each of the grids of the first grid dictionary.
- the Euclidean distance between each grid of the face image to be processed and each grid of the first grid dictionary is sequentially calculated, and the calculated distance is sorted, and the distance is selected to be the smallest.
- M grids in the first grid dictionary that is, each grid corresponding to the image of the face to be processed selects M fuzzy grids in the first grid dictionary.
- the Euclidean distance can be calculated by the following formula:
- the pixel of the tth grid of the face image to be processed The pixel of the i-th grid of the first grid dictionary.
- the grid of the face image to be processed in this embodiment is equal to the number of grids in the first grid dictionary.
- Step S233 acquiring M fuzzy meshes matching each mesh of the face image to be processed according to the calculated Euclidean distance.
- the distance is taken. Count down, then sort, and filter out the grids in the M first grid dictionary with the largest reciprocal.
- the distance is taken. Count down, and subtract the reciprocal of the distance with the constant A, and then sort to filter the meshes in the M first grid dictions with the largest difference.
- Step S24 according to the fuzzy grid, querying, in the second grid dictionary, a plurality of clear grids corresponding to the plurality of fuzzy grids, wherein the second grid dictionary is aligned to the second two-dimensional library according to the face template
- the second two-dimensional library is a two-dimensional face gallery of clear images created by the second three-dimensional face gallery obtained by the three-dimensional reconstruction method, and the blurred images are in one-to-one correspondence with the clear images.
- the grids of the corresponding positions of the first grid dictionary, the second grid dictionary, and the to-be-processed face image in this embodiment are the same in the face template. Therefore, after the plurality of fuzzy meshes corresponding to the mesh of the image of the face to be processed are queried in the first grid dictionary, the second grid dictionary may be queried according to the coordinates of the plurality of fuzzy meshes. To a plurality of clear meshes that correspond one-to-one with multiple fuzzy meshes.
- step S24 specifically includes the following steps:
- Step S241 acquiring coordinates of the blurred mesh on the face template.
- each mesh of the face image to be processed has M matching fuzzy meshes in the first dictionary, and the coordinates of the M fuzzy meshes on the face template are sequentially acquired, and Determine the location of the blurred mesh on the face template.
- Step S242 querying the clear grid corresponding to the fuzzy grid in the second grid dictionary according to the coordinates.
- the coordinates of the corresponding positions of the first grid dictionary, the second grid dictionary, and the face image to be processed in the embodiment are the same in the face template, the coordinates of the plurality of fuzzy grids correspond to the first The coordinates of a plurality of clear grids in the two grid dictionary, so that a plurality of clear grids can be determined based on the coordinates.
- step S25 the mesh of the face image to be processed is defuzzified according to the clear grid.
- the pixels of the M clear grids are processed, and the person to be processed is replaced.
- the corresponding mesh of the face image By repeating the above operations for all the meshes of the face image to be processed, the deblurring process of the face image to be processed can be realized.
- step S25 specifically includes the following steps:
- Step S251 acquiring pixels of a clear grid.
- the pixels of the grid of the second grid dictionary may be obtained by summing the pixel values of all the pixels in each grid and averaging; in addition, due to the image of the face to be processed, the first network
- the grid dictionary and the second grid dictionary are all aligned on the face template for meshing. Therefore, the number of meshes after the face image to be processed, the first grid dictionary, and the second grid dictionary are equal.
- N Use N to indicate the number of grids; the number of pixels in each grid is equal, so that all pixel points in the grid can also be used to form a pixel value vector, which is used to represent the pixels of the grid, ie
- the pixels of the grid of the second grid dictionary are represented by a pixel value vector composed of all pixels in the grid, that is,
- Step S252 the grid of the face image to be processed is processed such that the pixels of each grid of the face image to be processed are the sum of the pixels of the plurality of clear grids.
- the sum of the pixels of the M clear grids corresponding to each grid of the face image to be processed is calculated, and the corresponding image of the face to be processed is replaced by the sum of the pixels of the M clear grids.
- the pixels of the grid Therefore, by repeating the above operations for all the meshes of the face image to be processed in turn, the deblurring process of the face image to be processed can be realized.
- the present embodiment provides a face deblurring method for use in a face deblurring device. As shown in FIG. 6, the face deblurring method includes the following steps:
- Step S31 acquiring a first three-dimensional face gallery and a second three-dimensional face gallery obtained by using the three-dimensional reconstruction method, the first three-dimensional face gallery and the second three-dimensional face gallery respectively are two-dimensional blurred images and corresponding two-dimensional images Column expansion.
- a plurality of blurred images and corresponding two-dimensional columnar expansion images of the clear images are obtained, that is, the first three-dimensional face library and the second three-dimensional image. Face gallery.
- Step S32 configuring a posture parameter of the face image to be processed.
- the attitude parameter in this embodiment is an angle ( ⁇ x , ⁇ y , ⁇ z ) of the face image to be processed in a three-dimensional space; wherein ⁇ x is an offset angle of the to-be-processed face image in the x direction , ⁇ y is an offset angle of the to-be-processed face image in the y direction, and ⁇ z is an offset angle of the to-be-processed face image in the z direction.
- Step S33 establishing a corresponding first two-dimensional library and a second two-dimensional library in the first three-dimensional face gallery and the second three-dimensional face gallery respectively according to the posture parameter.
- the user configures the corresponding posture parameter according to the angle of the image to be processed in the three-dimensional space, and the first two-dimensional library corresponding to the posture parameter and the second from the first three-dimensional face library are obtained.
- the second two-dimensional library under the corresponding posture parameter is obtained in the three-dimensional face gallery.
- Step S34 acquiring a face image to be processed. It is the same as step S21 in Embodiment 2 and will not be described again.
- step S35 the image of the face to be processed is aligned onto the face template and meshed. The same as step S22 in the second embodiment, and the details are not described again.
- Step S36 Matching each mesh of the divided face image to be matched with the mesh of the first mesh dictionary, and obtaining a plurality of fuzzy meshes corresponding to each mesh of the to-be-processed face image, where
- the first grid dictionary is obtained by aligning the first two-dimensional library according to the face template, and the first two-dimensional library is a two-dimensional face gallery of the blurred image created by the first three-dimensional face gallery obtained by the three-dimensional reconstruction method. . It is the same as step S23 in Embodiment 2, and will not be described again.
- Step S37 according to the fuzzy grid, querying, in the second grid dictionary, a plurality of clear grids corresponding to the plurality of fuzzy grids, wherein the second grid dictionary is aligned to the second two-dimensional library according to the face template
- the second two-dimensional library is a two-dimensional face gallery of clear images created by the second three-dimensional face gallery obtained by the three-dimensional reconstruction method, and the blurred images are in one-to-one correspondence with the clear images. It is the same as step S24 in Embodiment 2, and will not be described again.
- step S38 the mesh of the face image to be processed is defuzzified according to the clear grid. It is the same as step S25 in Embodiment 2, and will not be described again.
- the face deblurring method provided in this embodiment can set the posture parameter of the face image to be processed in the space according to the user, so that the two-dimensional image dictionary under the corresponding posture parameter can be obtained from the histogram dictionary, and then the video can be processed.
- the face with the gesture under the monitoring scene is blurred.
- the present embodiment provides a face deblurring apparatus for performing the face deblurring method in Embodiments 1 to 3 of the present invention.
- the face deblurring device includes:
- a first acquiring unit 41 configured to acquire a to-be-processed face image
- a dividing unit 42 configured to align the image of the face to be processed onto the face template and perform the same on the face template Meshing.
- the matching unit 43 is configured to match each mesh of the divided face image to be processed with the mesh of the first mesh dictionary to obtain multiple fuzzy meshes corresponding to each mesh of the image to be processed.
- the first grid dictionary is obtained by aligning and dividing the first two-dimensional library according to the face template, and the first two-dimensional library is a two-dimensional face of the blurred image created by the three-dimensional face gallery obtained by the three-dimensional reconstruction method. Gallery.
- the query unit 44 is configured to query, according to the fuzzy mesh, a plurality of clear meshes corresponding to the plurality of fuzzy meshes in the second mesh dictionary, wherein the second mesh dictionary is a second two-dimensional library according to the After the face template is aligned and divided, the second two-dimensional library is a two-dimensional face gallery of clear images created by the three-dimensional face library obtained by the three-dimensional reconstruction method, and the blurred image corresponds to the clear image one by one.
- the processing unit 45 is configured to generate a clear image of the image of the face to be processed according to the clear grid that is queried.
- the face deblurring device provided in this embodiment can process face images of different postures and has a better face deblurring effect.
- the matching unit 43 includes:
- the second obtaining unit 431 is configured to respectively acquire each grid of the face image to be processed and pixels of each grid of the first grid dictionary.
- the calculating unit 432 is configured to calculate, according to the acquired pixels, a Euclidean distance of pixel similarity between each mesh of the to-be-processed face image and each mesh of the first grid dictionary.
- the third obtaining unit 433 is configured to acquire M fuzzy meshes that match each mesh of the to-be-processed face image according to the calculated Euclidean distance.
- the query unit 44 includes:
- the fourth obtaining unit 441 is configured to acquire coordinates of the blurred mesh on the face template.
- the query sub-unit 442 is configured to query a clear grid corresponding to the fuzzy grid in the second grid dictionary according to the coordinates.
- the processing unit 45 includes:
- the fifth obtaining unit 451 is configured to acquire pixels of a clear grid.
- the processing sub-unit 452 is configured to process the mesh of the face image to be processed such that the pixels of each mesh of the to-be-processed face image are the sum of the pixels of the plurality of clear meshes.
- the face deblurring device further includes:
- a sixth acquiring unit 46 configured to acquire a first three-dimensional face gallery and a second three-dimensional face gallery obtained by using a three-dimensional reconstruction method, wherein the first three-dimensional face gallery and the second three-dimensional face gallery respectively are a plurality of blurred images and corresponding A two-dimensional column expansion of a clear image.
- the configuration unit 47 is configured to configure a posture parameter of the face image to be processed.
- the establishing unit 48 is configured to establish a corresponding first two-dimensional library and a second two-dimensional library in the first three-dimensional face gallery and the second three-dimensional face gallery respectively according to the posture parameter.
- FIG. 9 is a schematic diagram showing the hardware structure of an image processing apparatus according to an embodiment of the present invention. As shown in FIG. 9, the apparatus includes one or more processors 51 and a memory 52. One processor 51 is taken as an example in FIG.
- the image processing apparatus may further include an image display (not shown) for comparing the processing results of the display image.
- the processor 51, the memory 52, and the image display may be connected by a bus or other means, as exemplified by a bus connection in FIG.
- the processor 51 can be a Central Processing Unit (CPU).
- the processor 51 can also be another general-purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or Other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., or a combination of the above various types of chips.
- the general purpose processor may be a microprocessor or the processor or any conventional processor or the like.
- the memory 52 is used as a non-transitory computer readable storage medium, and can be used for storing a non-transitory software program, a non-transitory computer executable program, and a module, such as a program instruction corresponding to the face deblurring method in the embodiment of the present invention. Module.
- the processor 51 executes various functional applications and numbers of the server by running non-transitory software programs, instructions, and modules stored in the memory 52. According to the processing, the face deblurring method in the above embodiment is implemented.
- the memory 52 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application required for at least one function; the storage data area may store data created according to the use of the face deblurring device, and the like.
- memory 52 can include high speed random access memory, and can also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device.
- memory 52 may optionally include memory remotely located relative to processor 51, which may be connected to the face deblurring device over a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
- the one or more modules are stored in the memory 52, and when executed by the one or more processors 51, the face deblurring method described in any one of Embodiments 1 to 3 is performed.
- the above product can perform the method provided by the embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.
- the technical details that are not described in detail in this embodiment, refer to the related description in the embodiment shown in FIG.
- the embodiment of the present invention further provides a non-transitory computer storage medium, where the computer storage medium stores computer executable instructions, which can execute the person described in any one of Embodiments 1 to 3. Face image deblurring method.
- the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk (Hard). Disk Drive, abbreviated as: HDD) or Solid-State Drive (SSD), etc.; the storage medium may also include a combination of the above types of memories.
- the storage medium may be a magnetic disk, an optical disk, a read only memory (ROM), or a random access memory (RAM).
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
本发明公开了一种人脸去模糊方法及装置，其中方法包括：获取待处理人脸图像；将待处理人脸图像对齐到人脸模板上，并对其进行网格划分；将划分后的待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到待处理人脸图像的每个网格对应的多个模糊网格，第一网格字典为对第一二维图库根据所述人脸模板对齐划分后得到的；根据模糊网格，在第二网格字典查询与所述多个模糊网格一一对应的多个清晰网格，第二网格字典为对第二二维图库根据人脸模板对齐划分后得到的，模糊图像与清晰图像一一对应；根据查询到的清晰网格，生成待处理人脸图像的清晰图像。本发明实施例提供的人脸去模糊方法能够处理不同姿态的人脸图像，具有较好的人脸去模糊效果。
Description
本发明涉及图像处理技术领域，具体涉及一种人脸去模糊方法及装置。
在人脸图像采集过程中，作为图像中的一类典型现象，采集到的图像往往会出现亮度失衡或者模糊的现象，从而会对图像质量有着极大的影响。尤其是随着各类缺乏稳定装置的只能终端和手持设备的广泛普及，所拍摄图像和视频包含模糊部分变得越来越常见。其中，影响图像清晰程度的因素有很多，如拍摄过程中的抖动、聚焦不准、曝光过度或不均以及摄像头和景物之间的相互移动，都会降低图像的质量，这一质量下降的过程称为图像的退化。
然而图像质量的不佳会对公安刑侦人员办案带来极大的不便，因为刑侦人员在办案时对人员追踪和人员身份的确认，很大程度上依赖各地各点的监控视频进行人工排查，或者使用人脸识别系统进行人脸比对。但是，各地各点的视频监控建设往往处于不同的阶段，会导致有些地方架设的场景比较欠缺，有些地方成像质量较差等等。例如，在有些视频监控场景下人脸成像会出现遮挡、模糊或姿态过大等等问题。
因此，现有技术中为解决上述技术问题，首先从待处理人脸图像中提取出模糊区域，然后通过相关算法对模糊区域进行处理，从模糊图像中恢复出隐式的清晰图像。
然而，上述技术方案中，需要从模糊图像中尽可能精确地检测出其中所包含的模糊区域，然后再通过模糊图像自身恢复出该模糊区域对应的清晰的图像，从而导致该去模糊化方法处理的效果不佳。
发明内容
有鉴于此，本发明实施例提供了一种人脸去模糊方法及装置，以解决现有技术中的人脸图像的去模糊化处理效果不佳的问题。
本发明第一方面提供了一种人脸去模糊方法，包括以下步骤：
获取待处理人脸图像；
将所述待处理人脸图像对齐到人脸模板上，并对其进行网格划分；
将划分后的所述待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到所述待处理人脸图像的每个网格对应的多个模糊网格，其中，所述第一网格字典为对第一二维图库根据所述人脸模板对齐划分后得到的，第一二维图库为利用三维重建方法得到的第一三维人脸图库建立的模糊图像的二维人脸图库；
根据所述模糊网格，在第二网格字典查询与所述多个模糊网格一一对应的多个清晰网格，其中，所述第二网格字典为对第二二维图库根据所述人脸模板对齐划分后得到的，第二二维图库为利用三维重建方法得到的第二三维人脸图库建立的清晰图像的二维人脸图库，所述模糊图像与所述清晰图像一一对应；
根据查询到的所述清晰网格，生成所述待处理人脸图像的清晰图像。
可选地，将划分后的所述待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到所述待处理人脸图像的每个网格对应的多个模糊网格，包括以下步骤：
分别获取所述待处理人脸图像的每个网格和所述第一网格字典的每个网格的像素；
根据获取到的像素计算所述待处理人脸图像的每个网格分别与所述第一网格字典的每个网格之间的像素的欧式距离；
根据计算得到的欧式距离获取与所述待处理人脸图像的每个网格匹配的M个模糊网格。
可选地，所述根据所述模糊网格，在第二网格字典查询与所述多个模糊网格一一对应的多个清晰网格，包括以下步骤：
获取所述模糊网格在人脸模板上的坐标；
根据所述坐标在所述第二网格字典中查询与所述模糊网格对应的清晰网格。
可选地，所述根据所述清晰网格，对所述待处理人脸图像的网格进行去模糊化处理，包括以下步骤：
获取所述清晰网格的像素；
对所述待处理人脸图像的网格进行处理，使得所述待处理人脸图像的每个网格的像素为所述多个清晰网格的像素的和。
可选地，在获取待处理人脸图像之前，包括以下步骤：
获取利用三维重建方法得到的第一三维人脸图库和第二三维人脸图库，所述第一三维人脸图库和所述第二三维人脸图库分别为若干模糊图像以及对应的清晰图像的二维柱状展开图；
配置所述待处理人脸图像的姿态参数；
根据所述姿态参数，分别在所述第一三维人脸图库和所述第二三维人脸图库中建立对应的第一二维图库和第二二维图库。
可选地，所述姿态参数为所述待处理人脸图像在三维空间内的角度(θx，θy，θz)；
其中，θx为所述待处理人脸图像在x方向上的偏移角度，θy为所述待处理人脸图像在y方向上的偏移角度，θz为所述待处理人脸图像在z方向上的偏移角度。
本发明第二方面提供了一种人脸去模糊装置，包括：
第一获取单元，用于获取待处理人脸图像；
划分单元，用于将待处理人脸图像对齐到人脸模板上，并对其进行网格划分；
匹配单元，用于将划分后的所述待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到所述待处理人脸图像的每个网格对应的多个模糊网格，其中，所述第一网格字典为对第一二维图库根据所述人脸模板对齐划分后得到的，第一二维图库为利用三维重建方法得到的三维人脸图库建立的模糊图像的二维人脸图库；
查询单元，用于根据所述模糊网格，在第二网格字典查询与所述多个模糊网格一一对应的多个清晰网格，其中，所述第二网格字典为对第二二维图库根据所述人脸模板对齐划分后得到的，第二二维图库为利用三维重建方法得到的三维人脸图库建立的清晰图像的二维人脸图库，所述模糊图像与所述清晰图像一一对应；
处理单元，用于根据查询到的所述清晰网格，生成所述待处理人脸图像的清晰图像。
可选地，所述匹配单元包括：
第二获取单元，用于分别获取所述待处理人脸图像的每个网格和所述第一网格字典的每个网格的像素；
计算单元，用于根据获取到的像素计算所述待处理人脸图像的每个网格分别与所述第一网格字典的每个网格之间的像素相似度的欧式距离；
第三获取单元，用于根据计算得到的欧式距离获取与所述待处理人脸图像的每个网格匹配的M个模糊网格。
本发明第三方面提供了一种图像处理装置，包括至少一个处理器；以及与所述至少一个处理器通信连接的存储器；其中，所述存储器存储有可被所述一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器执行本发明第一方面中任一项所述的人脸去模糊方法。
本发明第四方面提供了一种非暂态计算机可读存储介质，所述非暂态计算机可读存储介质存储计算机指令，所述计算机指令用于使所述计算机执行第一方面或者第一方面的任意一种可选方式中所述的人脸去模糊方法。
本发明第五方面提供了一种计算机程序产品，所述计算机程序产品包括存储在非暂态计算机可读存储介质上的计算程序，所述计算机程序包括程序指令，当所述程序指令被计算机执行时，使所述计算机执行第一方面或者第一方面的任意一种可选方式中所述的人脸去模糊。
本发明提供的技术方案，具有如下优点：
1.本发明实施例提供的人脸模糊方法，包括以下步骤：获取待处理人脸图像；将所述待处理人脸图像对齐到人脸模板上，并对其进行网格划分；将划分后的所述待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到所述待处理人脸图像的每个网格对应的多个模糊网格，其中，所述第一网格字典为对第一二维图库根据所述人脸模板对齐划分后得到的，第一二维图库为利用三维重建方法得到的第一三维人脸图库建立的模糊图像的二维人脸图库；根据所述模糊网格，在第二网格字典查询与所述多个模糊网格一一对应的多个清晰网格，其中，所述第二网格字典为对第二二维图库根据所述人脸模板对齐划分后得到的，第二二维图库为利用三维重建方法得到的第二三维人脸图库建立的清晰图像的二维人脸图库，所述模糊图像与所述清晰图像一一对应；根据查询到的所述清晰网格，生成所述待处理人脸图像的清晰图像。本发明实施例提供的人脸去模糊方法能够处理不同姿态的人脸图像，具有较好的人脸去模糊效果。
2.本发明实施例提供的人脸去模糊方法，其中，根据所述清晰网格，对所述待处理人脸图像的网格进行去模糊化处理，包括以下步骤：获取所述清晰网格的像素；对所述待处理人脸图像的网格进行处理，使得所述待处理人脸图像的每个网格的像素为所述多个清晰网格的像素的和。本发明实施例在用清晰网格替换模糊网格的过程中，对于固定位置的网格像素，选用不同清晰人脸上的同一位置的网格像素加权计算得到的网格像素替换原来模糊网格像素，具有较好的人脸去模糊效果。
3.本发明实施例提供的人脸去模糊方法，其中，在获取待处理人脸图像之前，包括以下步骤：获取利用三维重建方法得到的第一三维人脸图库和第二三维人脸图库，所述第一三维人脸图库和所述第二三维人脸图库分别为若干模糊图像以及对应的清晰图像的二维柱状展开图；配置所述待处理人脸图像的姿态参数；根据所述姿态参数，分别在所述第一三维人脸图库和所述第二三维人脸图库中建立对应的第一二维图库和第二二维图库。本发明实施例提供的人脸去模糊方法，能够根据用户设定待处理人脸图像在空间的姿态参数，从而可从柱状图字典中获取对应的姿态参数下的二维
图像字典，进而可以处理视频监控场景下含姿态的人脸去模糊处理。
4.本实施例提供的人脸去模糊装置，包括：第一获取单元，用于获取待处理人脸图像；划分单元，用于将待处理人脸图像对齐到人脸模板上，并对其进行网格划分；匹配单元，用于将划分后的所述待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到所述待处理人脸图像的每个网格对应的多个模糊网格，其中，所述第一网格字典为对第一二维图库根据所述人脸模板对齐划分后得到的，第一二维图库为利用三维重建方法得到的三维人脸图库建立的模糊图像的二维人脸图库；查询单元，用于根据所述模糊网格，在第二网格字典查询与所述多个模糊网格一一对应的多个清晰网格，其中，所述第二网格字典为对第二二维图库根据所述人脸模板对齐划分后得到的，第二二维图库为利用三维重建方法得到的三维人脸图库建立的清晰图像的二维人脸图库，所述模糊图像与所述清晰图像一一对应；处理单元，用于根据查询到的所述清晰网格，生产所述待处理人脸图像的清晰图像。本发明实施例提供的人脸去模糊装置能够处理不同姿态的人脸图像，具有较好的人脸去模糊效果。
通过参考附图会更加清楚的理解本发明的特征和优点，附图是示意性的而不应理解为对本发明进行任何限制，在附图中：
图1示出了三维人脸重建中正面人脸转其他姿态人脸的示意图；
图2示出了人脸的三维形状模型；
图3示出了三维人脸纹理的二维柱状展开图；
图4示出了本发明实施例1中人脸去模糊方法的一个具体示意的流程图；
图5示出了本发明实施例2中人脸去模糊方法的一个具体示意的流程图；
图6示出了本发明实施例3中人脸去模糊方法的一个具体示意的流程图；
图7示出了本发明实施例4中人脸去模糊装置的一个具体示意的结构图；
图8示出了本发明实施例4中人脸去模糊装置的又一个具体示意的结构图；
图9示出了本发明实施例5中图像处理装置的一个具体示意的结构图。
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本发明实施例中，三维人脸重建的过程是利用三维人脸模型将正面人脸图像转换为任意其他角度的人脸图像，具体过程如图1所示，即对于一张正面人脸图像只要给定姿态参数就能得到相应姿态下的人脸图像。其中，三维人脸模型中包含以下四方面的信息：
(1)一张正面人脸三维形状S＝(x1，y1，z1，...，xn，yn，zn)T，其中n表示形状顶点个数；
(2)一张正面人脸的三维纹理T＝(r1，g1，b1，...，rm，gm，bm)T，其中m表示纹理图像像素个数；
(3)以上正面人脸的形状顶点在纹理图上的对应关系，即二维形状的每一个形状点都能对应到纹理像素值，如图2所示的示意人脸的三维形状模型；
(4)其它不在模型中形状点像素值通过那些周围在模型中的形状顶点上的像素插值得到。
对于任意一张二维正面人脸图像，正面人脸重建任意角度人脸的过程如下：
步骤1：利用ASM定位出68个人脸关键点；
步骤2：利用68个关键点将当前人脸二维形状对齐到模型中的三维形状，因为是正面人脸，Z方向的旋转角为0，所以需要参数(θx，θy，θz，Δx，Δy，Δz，s)，其中，θx，θy，θz分别表示x，y，z三个方向的旋转角度、Δx，Δy，Δz分别表示x，y，z三个方向的平移和s表示缩放因子；
步骤3：再利用步骤2计算出来的旋转、平移和缩放参数对上述描述形状S每个顶点进行旋转、平移和缩放，即当前形状对齐到模型中的标准形状；
步骤4：最后利用步骤(2)和步骤(3)中的信息，结合Kriging插值得到如图3所示为完整的三维人脸纹理图的二维柱状展开图；
步骤5：当用户输入将当前正面人脸旋转的角度参数(θx，θy，θz)时，对齐后的形状顶点进行旋转(步骤2的逆过程)；
步骤6：根据步骤5计算的参数，结合步骤4生成的三维人脸纹理图的二维柱状展开图，利用Kriging插值得到目标图像。
引入三维人脸重建进行任意姿态的人脸处理时，只需要事前建立多张人脸的三维纹理图库。当客户配置姿态参数时，可以变换得到该姿态下的二维纹理图库，利用该二维纹理图库在线训练不同的人脸处理模型就满足任意姿态的不同图像处理。
实施例1
本施例提供一种人脸去模糊方法，用于人脸去模糊装置中。如图4所示，该人脸去模糊方法包括以下步骤：
步骤S11，获取待处理人脸图像。
本实施例中的人脸图像可以是事先存储在人脸去模糊装置中的，或人脸去模糊装置实时从外界获取的图像，也可以是人脸去模糊装置从一段视频中提取的图像。
步骤S12，将待处理人脸图像对齐到人脸模板上，并对其进行网格划分。
由于人脸拥有比较固定的结构，如眉毛、眼睛、鼻子、嘴巴，单独看眼睛、眉毛、鼻子和嘴巴等器官差别不大，再将各个器官划分成很多纹理
小块，差别就更有限。本发明实施例基于此，对人脸图像划分为小的网格，每个器官由多个网格组成。
本实施例中的人脸模板，即为待处理人脸图像进行划分的基准。在对待处理人脸图像进行网格划分时，先将待处理人脸图像进行五官的对齐和尺寸归一化处理，即人脸的眼睛、眉毛、鼻子、嘴巴大体与人脸模板的对应器官在同一位置，从而能够保证划分的标准化和精确性。
本实施例中，可以将待处理人脸图像划分为m行n列的网格，其中，m为垂直方向上的网格数量，n为水平方向上的网格数量。m、n的具体数值可以根据实际对处理后的待处理人脸图像的精度进行设定，可以相等，也可以不相等。
步骤S13，将划分后的待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到待处理人脸图像的每个网格对应的多个模糊网格，其中，第一网格字典为对第一二维图库根据人脸模板对齐划分后得到的，第一二维图库为利用三维重建方法得到的第一三维人脸图库建立的模糊图像的二维人脸图库。
本实施例中，第一三维人脸图库为S张脸的模糊图像的人脸字典库，即利用上述的三维人脸重建方法获得相应的模糊图像的二维柱状展开图；可以根据待处理人脸图像的相对于正面图像的偏转角度，从该二维柱状展开图中提取出第一二维图库。第一网格字典为对第一二维图库根据人脸模板对齐划分后得到的；其中，这S张脸的模糊图像在对齐时进行了五官的对齐和尺寸归一化，即每张人脸的眼睛、眉毛、鼻子、嘴巴大体在画面中的同一位置，从而使得第一网格字典中的网格与待处理人脸图像的对应位置的网格在人脸模板上的坐标相同。
本实施例通过使用第一网格字典中的多个模糊网格与待处理人脸图像的一个网格匹配，从而提高了匹配的精确性，达到提高处理后的待处理人脸图像的清晰度的目的。
步骤S14，根据模糊网格，在第二网格字典查询与多个模糊网格一一对应的多个清晰网格，其中，第二网格字典为对第二二维图库根据人脸模板
对齐划分后得到的，第二二维图库为利用三维重建方法得到的第二三维人脸图库建立的清晰图像的二维人脸图库，模糊图像与清晰图像一一对应。
本实施例中，第二三维人脸图库为S张脸的清晰图像的人脸字典库，即利用上述的三维人脸重建方法获得相应的清晰图像的二维柱状展开图；可以根据待处理人脸图像的相对于正面图像的偏转角度，从该二维柱状展开图中提取出第二二维图库。第二网格字典为对第二二维图库根据人脸模板对齐划分后得到的，即第一网格字典、第二网格字典以及待处理人脸图像对应位置的网格在人脸模板上的坐标相同。
由于上述三者之间的对应位置的坐标相同，从而为在第二网格字典中查询与多个模糊网格一一对应的多个清晰网格提供了便利。
需要说明的是，本发明实施例中所述的模糊图像和清晰图像以及后面所提到的模糊网格和清晰网格都是相对的，例如，清晰图像表示人眼可快速辨识的图像，其具体可以通过图像某些参数(例如像素值)来界定，模糊图像同理。
步骤S15，根据查询到的清晰网格，生成待处理人脸图像的清晰图像。
本实施例中，可以用查询到的清晰网格直接取代待处理人脸图像中的网格，也可以用查询到的清晰网格的像素处理后，取代待处理人脸图像中的网格像素。
实施例2
本施例提供一种人脸去模糊方法，用于人脸去模糊装置中。如图5所示，该人脸去模糊方法包括以下步骤：
步骤S21，获取待处理人脸图像。与实施例1中的步骤S11相同，不再赘述。
步骤S22，将待处理人脸图像对齐到人脸模板上，并对其进行网格划分。与实施例1中的步骤S12相同，不再赘述。
步骤S23，将划分后的待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到待处理人脸图像的每个网格对应的多个模糊网格，其中，
第一网格字典为对第一二维图库根据人脸模板对齐划分后得到的，第一二维图库为利用三维重建方法得到的第一三维人脸图库建立的模糊图像的二维人脸图库。
本实施例通过计算待处理人脸图像的网格与第一网格字典的网格的像素之间的欧式距离，并根据该欧式距离找出与待处理人脸图像的网格匹配的第一网格字典中的网格。
作为本实施例的一种可选实施方式，步骤S23具体包括以下步骤：
步骤S231，分别获取待处理人脸图像的每个网格和第一网格字典的每个网格的像素。
本实施例中，待处理人脸图像的网格的像素和第一网格字典的网格的像素是通过可以对每个网格中所有像素点的像素值求和后取平均值得到；此外，由于待处理人脸图像与第一网格字典都是对齐到人脸模板上进行网格划分的，因此，待处理人脸图像与第一网格字典划分出的网格数量是想等的，用N表示网格的数量；每个网格内的像素点的个数是相等的，从而也可以利用网格中的所有像素点组成的像素值向量，用该向量表示网格的像素，即待处理人脸图像的网格的像素表示为
其中，t为待处理人脸图像的第t个网格，t＝1至N；第一网格字典的网格的像素表示为
其中，i为第一网格字典中的第i个网格，i＝1至N。
作为本实施例的一种可选实施方式，待处理人脸图像的网格的像素和第一网格字典的网格的像素是通过网格内所有像素点组成的像素值向量表示的。
步骤S232，根据获取到的像素计算待处理人脸图像的每个网格分别与第一网格字典的每个网格之间的像素的欧式距离。
本实施例中，依次计算待处理人脸图像的每个网格分别与第一网格字典的每个网格之间的像素的欧式距离，对计算得到的距离进行排序，筛选出距离最小的M个第一网格字典中的网格；即对应于待处理人脸图像的每个网格在第一网格字典中筛选出M个模糊网格。其中，M的取值可以为10-20的任意值，作为本实施例的一种可选实施方式，M＝10，该数值在保
证较精确的筛选效果的前提下，还能够简化计算过程。其中，欧式距离可以通过如下公式计算：
其中，
为待处理人脸图像的第t个网格的像素，
为第一网格字典的第i个网格的像素。
本实施例中的待处理人脸图像的网格与第一网格字典中的网格数量相等，在计算过程中，依次计算当t＝1至N时，对应的欧式距离值，即：
当t＝1时，计算
i＝1至N；其中，
为待处理人脸图像的第一个网格的像素；
当t＝2时，计算
i＝1至N；其中，
为待处理人脸图像的第二个网格的像素；
当t＝N时，计算
i＝1至N；其中，
为待处理人脸图像的第N个网格的像素。
步骤S233，根据计算得到的欧式距离获取与待处理人脸图像的每个网格匹配的M个模糊网格。
作为本实施例的一种可选实施方式，在计算出待处理人脸图像的每个网格分别与第一网格字典的每个网格之间的像素的欧式距离之后，取该距离的倒数，然后进行排序，筛选出倒数最大的M个第一网格字典中的网格。
作为本实施例的一种可选实施方式，在计算出待处理人脸图像的每个网格分别与第一网格字典的每个网格之间的像素的欧式距离之后，取该距离的倒数，并用常数A减去该距离的倒数，然后进行排序，筛选出差值最大的M个第一网格字典中的网格。本实施例中，A＝1，能够达到在提高人脸图像的去模糊精度的基础上，简化计算。
步骤S24，根据模糊网格，在第二网格字典查询与多个模糊网格一一对应的多个清晰网格，其中，第二网格字典为对第二二维图库根据人脸模板对齐划分后得到的，第二二维图库为利用三维重建方法得到的第二三维人脸图库建立的清晰图像的二维人脸图库，模糊图像与清晰图像一一对应。
本实施例中的第一网格字典、第二网格字典以及待处理人脸图像的对应位置的网格在人脸模板中的坐标相同。因此，当在第一网格字典中，查询到与待处理人脸图像的网格对应的多个模糊网格后，可以根据该多个模糊网格的坐标，在第二网格字典中查询到与多个模糊网格一一对应的多个清晰网格。
作为本实施例的一种可选实施方式，步骤S24具体包括以下步骤：
步骤S241，获取模糊网格在人脸模板上的坐标。
本实施例中，待处理人脸图像的每个网格在第一字典中都有M个匹配的模糊网格，依次获取到该M个模糊网格在人脸模板上的坐标，即可在人脸模板上确定模糊网格的位置。
步骤S242，根据坐标在第二网格字典中查询与模糊网格对应的清晰网格。
由于本实施例中第一网格字典、第二网格字典以及待处理人脸图像的对应位置的网格在人脸模板中的坐标相同，因此，多个模糊网格的坐标即对应到第二网格字典中的多个清晰网格的坐标，从而根据该坐标能够确定多个清晰网格。
步骤S25，根据清晰网格，对待处理人脸图像的网格进行去模糊化处理。
本实施例中，在第二网格字典中查询到与待处理人脸图像的每个网格对应的M个清晰网格后，对M个清晰网格的像素进行处理后，替换待处理人脸图像的对应的网格。对待处理人脸图像的所有网格重复上述操作，即可实现待处理人脸图像的去模糊处理。
作为本实施例的一种可选实施方式，步骤S25具体包括以下步骤：
步骤S251，获取清晰网格的像素。
本实施例中，第二网格字典的网格的像素可以是通过对每个网格中所有像素点的像素值求和后取平均值得到；此外，由于待处理人脸图像、第一网格字典以及第二网格字典都是对齐到人脸模板上进行网格划分的，因此，待处理人脸图像、第一网格字典以及第二网格字典划分后的网格数量是相等的，用N表示网格的数量；每个网格内的像素点的个数是相等的，
从而也可以利用网格中的所有像素点组成像素值向量，用该向量表示网格的像素，即第二网格字典的网格的像素表示为
其中j为第二网格字典中的第j个网格，j＝1至N。
作为本实施例的一种可选实施方式，第二网格字典的网格的像素是通过网格内所有像素点组成的像素值向量表示的，即
步骤S252，对待处理人脸图像的网格进行处理，使得待处理人脸图像的每个网格的像素为多个清晰网格的像素的和。
本实施例中，通过计算得到待处理人脸图像的每个网格对应的M个清晰网格的像素的和，并用该M个清晰网格的像素的和替换对应的待处理人脸图像的网格的像素。因此，依次对待处理人脸图像的所有网格重复上述操作，即可实现对待处理人脸图像的去模糊处理。
实施例3
本施例提供一种人脸去模糊方法，用于人脸去模糊装置中。如图6所示，该人脸去模糊方法包括以下步骤：
步骤S31，获取利用三维重建方法得到的第一三维人脸图库和第二三维人脸图库，第一三维人脸图库和第二三维人脸图库分别为若干模糊图像以及对应的清晰图像的二维柱状展开图。
利用本发明中的三维重建方法分别对若干模糊图像以及对应的清晰图像进行处理后，得到若干模糊图像以及对应的清晰图像的二维柱状展开图，即为第一三维人脸图库和第二三维人脸图库。
步骤S32，配置待处理人脸图像的姿态参数。
本实施例中的姿态参数为待处理人脸图像在三维空间内的角度(θx，θy，θz)；其中，θx为所述待处理人脸图像在x方向上的偏移角度，θy为所述待处理人脸图像在y方向上的偏移角度，θz为所述待处理人脸图像在z方向上的偏移角度。
步骤S33，根据姿态参数，分别在第一三维人脸图库和第二三维人脸图库中建立对应的第一二维图库和第二二维图库。
本实施例中，用户根据待处理人脸图像在三维空间内的角度，配置对应的姿态参数，即可从第一三维人脸图库中获取对应姿态参数下的第一二维图库和从第二三维人脸图库中获取对应姿态参数下的第二二维图库。
步骤S34，获取待处理人脸图像。与实施例2中的步骤S21相同，不再赘述。
步骤S35，将待处理人脸图像对齐到人脸模板上，并对其进行网格划分。与实施例2中的步骤S22相同，不再赘述。
步骤S36，将划分后的待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到待处理人脸图像的每个网格对应的多个模糊网格，其中，第一网格字典为对第一二维图库根据人脸模板对齐划分后得到的，第一二维图库为利用三维重建方法得到的第一三维人脸图库建立的模糊图像的二维人脸图库。与实施例2中的步骤S23相同，不再赘述。
步骤S37，根据模糊网格，在第二网格字典查询与多个模糊网格一一对应的多个清晰网格，其中，第二网格字典为对第二二维图库根据人脸模板对齐划分后得到的，第二二维图库为利用三维重建方法得到的第二三维人脸图库建立的清晰图像的二维人脸图库，模糊图像与清晰图像一一对应。与实施例2中的步骤S24相同，不再赘述。
步骤S38，根据清晰网格，对待处理人脸图像的网格进行去模糊化处理。与实施例2中的步骤S25相同，不再赘述。
本实施例提供的人脸去模糊方法，能够根据用户设定待处理人脸图像在空间的姿态参数，从而可从柱状图字典中获取对应的姿态参数下的二维图像字典，进而可以处理视频监控场景下含姿态的人脸去模糊处理。
实施例4
本施例提供一种人脸去模糊装置，用于执行本发明实施例1至实施例3中的人脸去模糊方法。如图7所示，该人脸去模糊装置包括：
第一获取单元41，用于获取待处理人脸图像；
划分单元42，用于将待处理人脸图像对齐到人脸模板上，并对其进行
网格划分。
匹配单元43，用于将划分后的待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到待处理人脸图像的每个网格对应的多个模糊网格，其中，第一网格字典为对第一二维图库根据人脸模板对齐划分后得到的，第一二维图库为利用三维重建方法得到的三维人脸图库建立的模糊图像的二维人脸图库。
查询单元44，用于根据模糊网格，在第二网格字典查询与多个模糊网格一一对应的多个清晰网格，其中，第二网格字典为对第二二维图库根据所述人脸模板对齐划分后得到的，第二二维图库为利用三维重建方法得到的三维人脸图库建立的清晰图像的二维人脸图库，模糊图像与清晰图像一一对应。
处理单元45，用于根据查询到的清晰网格，生产待处理人脸图像的清晰图像。
本实施例提供的人脸去模糊装置能够处理不同姿态的人脸图像，具有较好的人脸去模糊效果。
作为本实施例的一种可选实施方式，如图8所示，匹配单元43包括：
第二获取单元431，用于分别获取待处理人脸图像的每个网格和第一网格字典的每个网格的像素。
计算单元432，用于根据获取到的像素计算待处理人脸图像的每个网格分别与第一网格字典的每个网格之间的像素相似度的欧式距离。
第三获取单元433，用于根据计算得到的欧式距离获取与待处理人脸图像的每个网格匹配的M个模糊网格。
作为本实施例的一种可选实施方式，如图8所示，查询单元44包括：
第四获取单元441，用于获取模糊网格在人脸模板上的坐标。
查询子单元442，用于根据坐标在第二网格字典中查询与模糊网格对应的清晰网格。
作为本实施例的一种可选实施方式，如图8所示，处理单元45包括：
第五获取单元451，用于获取清晰网格的像素。
处理子单元452，用于对待处理人脸图像的网格进行处理，使得待处理人脸图像的每个网格的像素为多个清晰网格的像素的和。
作为本实施例的一种可选实施方式，如图8所示，该人脸去模糊装置还包括：
第六获取单元46，用于获取利用三维重建方法得到的第一三维人脸图库和第二三维人脸图库，第一三维人脸图库和第二三维人脸图库分别为若干模糊图像以及对应的清晰图像的二维柱状展开图。
配置单元47，用于配置待处理人脸图像的姿态参数。
建立单元48，用于根据所述姿态参数，分别在第一三维人脸图库和第二三维人脸图库中建立对应的第一二维图库和第二二维图库。
实施例5
图9是本发明实施例提供的图像处理装置的硬件结构示意图，如图9所示，该装置包括一个或多个处理器51以及存储器52，图9中以一个处理器51为例。
该图像处理装置还可以包括：图像显示器(未示出)，用于对比显示图像的处理结果。处理器51、存储器52和图像显示器可以通过总线或者其他方式连接，图9中以通过总线连接为例。
处理器51可以为中央处理器(Central Processing Unit，CPU)。处理器51还可以为其他通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field-Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等芯片，或者上述各类芯片的组合。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
存储器52作为一种非暂态计算机可读存储介质，可用于存储非暂态软件程序、非暂态计算机可执行程序以及模块，如本发明实施例中的人脸去模糊方法对应的程序指令/模块。处理器51通过运行存储在存储器52中的非暂态软件程序、指令以及模块，从而执行服务器的各种功能应用以及数
据处理，即实现上述实施例中的人脸去模糊方法。
存储器52可以包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需要的应用程序；存储数据区可存储根据人脸去模糊装置的使用所创建的数据等。此外，存储器52可以包括高速随机存取存储器，还可以包括非暂态存储器，例如至少一个磁盘存储器件、闪存器件、或其他非暂态固态存储器件。在一些实施例中，存储器52可选包括相对于处理器51远程设置的存储器，这些远程存储器可以通过网络连接至人脸去模糊装置。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
所述一个或者多个模块存储在所述存储器52中，当被所述一个或者多个处理器51执行时，执行实施例1至实施例3中任一项所述的人脸去模糊方法。
上述产品可执行本发明实施例所提供的方法，具备执行方法相应的功能模块和有益效果。未在本实施例中详尽描述的技术细节，具体可参见如图1所示的实施例中的相关描述。
实施例6
本发明实施例还提供了一种非暂态计算机存储介质，所述计算机存储介质存储有计算机可执行指令，该计算机可执行指令可执行实施例1至实施例3中任一项所述的人脸图像去模糊方法。其中，所述存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)、随机存储记忆体(Random Access Memory，RAM)、快闪存储器(Flash Memory)、硬盘(Hard Disk Drive，缩写：HDD)或固态硬盘(Solid-State Drive，SSD)等；所述存储介质还可以包括上述种类的存储器的组合。
本领域技术人员可以理解，实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，所述的存储介质可为磁碟、光盘、只读存储记忆体(ROM)或随机存储记忆体(RAM)等。
虽然结合附图描述了本发明的实施例，但是本领域技术人员可以在不脱离本发明的精神和范围的情况下作出各种修改和变型，这样的修改和变型均落入由所附权利要求所限定的范围之内。
Claims (10)
- 一种人脸去模糊方法，其特征在于，包括以下步骤：获取待处理人脸图像；将所述待处理人脸图像对齐到人脸模板上，并对其进行网格划分；将划分后的所述待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到所述待处理人脸图像的每个网格对应的多个模糊网格，其中，所述第一网格字典为对第一二维图库根据所述人脸模板对齐划分后得到的，第一二维图库为利用三维重建方法得到的第一三维人脸图库建立的模糊图像的二维人脸图库；根据所述模糊网格，在第二网格字典查询与所述多个模糊网格一一对应的多个清晰网格，其中，所述第二网格字典为对第二二维图库根据所述人脸模板对齐划分后得到的，第二二维图库为利用三维重建方法得到的第二三维人脸图库建立的清晰图像的二维人脸图库，所述模糊图像与所述清晰图像一一对应；根据查询到的所述清晰网格，生成所述待处理人脸图像的清晰图像。
- 根据权利要求1所述的人脸去模糊方法，其特征在于，将划分后的所述待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到所述待处理人脸图像的每个网格对应的多个模糊网格，包括以下步骤：分别获取所述待处理人脸图像的每个网格和所述第一网格字典的每个网格的像素；根据获取到的像素计算所述待处理人脸图像的每个网格分别与所述第一网格字典的每个网格之间的像素的欧式距离；根据计算得到的欧式距离获取与所述待处理人脸图像的每个网格匹配的M个模糊网格。
- 根据权利要求1或2所述的人脸去模糊方法，其特征在于，所述根据所述模糊网格，在第二网格字典查询与所述多个模糊网格一一对应的多个清晰网格，包括以下步骤：获取所述模糊网格在人脸模板上的坐标；根据所述坐标在所述第二网格字典中查询与所述模糊网格对应的清晰网格。
- 根据权利要求1至3中任一项所述的人脸去模糊方法，其特征在于，所述根据所述清晰网格，对所述待处理人脸图像的网格进行去模糊化处理，包括以下步骤：获取所述清晰网格的像素；对所述待处理人脸图像的网格进行处理，使得所述待处理人脸图像的每个网格的像素为所述多个清晰网格的像素的和。
- 根据权利要求1至4中任一项所述的人脸去模糊方法，其特征在于，在获取待处理人脸图像之前，包括以下步骤：获取利用三维重建方法得到的第一三维人脸图库和第二三维人脸图库，所述第一三维人脸图库和所述第二三维人脸图库分别为若干模糊图像以及对应的清晰图像的二维柱状展开图；配置所述待处理人脸图像的姿态参数；根据所述姿态参数，分别在所述第一三维人脸图库和所述第二三维人脸图库中建立对应的第一二维图库和第二二维图库。
- 根据权利要求5所述的人脸去模糊方法，其特征在于，所述姿态参数为所述待处理人脸图像在三维空间内的角度(θx，θy，θz)；其中，θx为所述待处理人脸图像在x方向上的偏移角度，θy为所述待处理人脸图像在y方向上的偏移角度，θz为所述待处理人脸图像在z方向上的偏移角度。
- 一种人脸去模糊装置，其特征在于，包括：第一获取单元，用于获取待处理人脸图像；划分单元，用于将待处理人脸图像对齐到人脸模板上，并对其进行网格划分；匹配单元，用于将划分后的所述待处理人脸图像的每个网格与第一网格字典的网格进行匹配，得到所述待处理人脸图像的每个网格对应的多个模糊网格，其中，所述第一网格字典为对第一二维图库根据所述人脸模板 对齐划分后得到的，第一二维图库为利用三维重建方法得到的三维人脸图库建立的模糊图像的二维人脸图库；查询单元，用于根据所述模糊网格，在第二网格字典查询与所述多个模糊网格一一对应的多个清晰网格，其中，所述第二网格字典为对第二二维图库根据所述人脸模板对齐划分后得到的，第二二维图库为利用三维重建方法得到的三维人脸图库建立的清晰图像的二维人脸图库，所述模糊图像与所述清晰图像一一对应；处理单元，用于根据查询到的所述清晰网格，生成所述待处理人脸图像的清晰图像。
- 根据权利要求7所述的人脸去模糊装置，其特征在于，所述匹配单元包括：第二获取单元，用于分别获取所述待处理人脸图像的每个网格和所述第一网格字典的每个网格的像素；计算单元，用于根据获取到的像素计算所述待处理人脸图像的每个网格分别与所述第一网格字典的每个网格之间的像素相似度的欧式距离；第三获取单元，用于根据计算得到的欧式距离获取与所述待处理人脸图像的每个网格匹配的M个模糊网格。
- 一种图像处理装置，其特征在于，包括至少一个处理器；以及与所述至少一个处理器通信连接的存储器；其中，所述存储器存储有可被所述一个处理器执行的指令，所述指令被所述至少一个处理器执行，以使所述至少一个处理器执行权利要求1-6中任一项所述的人脸去模糊方法。
- 一种非暂态计算机可读存储介质，其特征在于，所述非暂态计算机可读存储介质存储计算机指令，所述计算机指令用于使计算机执行权利要求1至6中任一项所述的人脸去模糊方法。
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| EP17923783.9A EP3598385B1 (en) | 2017-08-31 | 2017-12-19 | Face deblurring method and device |
| US16/654,779 US20200051228A1 (en) | 2017-08-31 | 2019-10-16 | Face Deblurring Method and Device |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710774351.1A CN107563978A (zh) | 2017-08-31 | 2017-08-31 | 人脸去模糊方法及装置 |
| CN201710774351.1 | 2017-08-31 |
Related Child Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| US16/654,779 Continuation US20200051228A1 (en) | 2017-08-31 | 2019-10-16 | Face Deblurring Method and Device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2019041660A1 true WO2019041660A1 (zh) | 2019-03-07 |
Family
ID=60977644
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2017/117166 WO2019041660A1 (zh) | 2017-08-31 | 2017-12-19 | 人脸去模糊方法及装置 |
Country Status (4)
| Country | Link |
|---|---|
| US (1) | US20200051228A1 (zh) |
| EP (1) | EP3598385B1 (zh) |
| CN (1) | CN107563978A (zh) |
| WO (1) | WO2019041660A1 (zh) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3340624B1 (en) * | 2016-12-20 | 2019-07-03 | Axis AB | Encoding a privacy masked image |
| CN107563978A (zh) * | 2017-08-31 | 2018-01-09 | 苏州科达科技股份有限公司 | 人脸去模糊方法及装置 |
| CN110020578A (zh) * | 2018-01-10 | 2019-07-16 | 广东欧珀移动通信有限公司 | 图像处理方法、装置、存储介质及电子设备 |
| WO2020056769A1 (en) * | 2018-09-21 | 2020-03-26 | Intel Corporation | Method and system of facial resolution upsampling for image processing |
| CN109345486A (zh) * | 2018-10-24 | 2019-02-15 | 中科天网（广东）科技有限公司 | 一种基于自适应网格变形的人脸图像去模糊方法 |
| CN109903233B (zh) * | 2019-01-10 | 2021-08-03 | 华中科技大学 | 一种基于线性特征的联合图像复原和匹配方法及系统 |
| CN111340722B (zh) * | 2020-02-20 | 2023-05-26 | Oppo广东移动通信有限公司 | 图像处理方法、处理装置、终端设备及可读存储介质 |
| US11341701B1 (en) * | 2021-05-06 | 2022-05-24 | Motorola Solutions, Inc | Method and apparatus for producing a composite image of a suspect |
| CN113344832A (zh) * | 2021-05-28 | 2021-09-03 | 杭州睿胜软件有限公司 | 图像处理方法及装置、电子设备和存储介质 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104484803A (zh) * | 2014-11-24 | 2015-04-01 | 苏州福丰科技有限公司 | 基于神经网络的三维人脸识别的手机支付方法 |
| CN105678252A (zh) * | 2016-01-05 | 2016-06-15 | 安阳师范学院 | 依据人脸三角网格自适应细分和高斯小波的迭代插值方法 |
| WO2017131343A1 (en) * | 2016-01-30 | 2017-08-03 | Samsung Electronics Co., Ltd. | A device for and method of enhancing quality of an image |
| CN107563978A (zh) * | 2017-08-31 | 2018-01-09 | 苏州科达科技股份有限公司 | 人脸去模糊方法及装置 |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4818053B2 (ja) * | 2006-10-10 | 2011-11-16 | 株式会社東芝 | 高解像度化装置および方法 |
| EP2291822A1 (en) * | 2008-05-21 | 2011-03-09 | Koninklijke Philips Electronics N.V. | Image resolution enhancement |
| CN101364302A (zh) * | 2008-09-28 | 2009-02-11 | 西安理工大学 | 一种散焦模糊图像的清晰化处理方法 |
| CN101477684B (zh) * | 2008-12-11 | 2010-11-10 | 西安交通大学 | 一种利用位置图像块重建的人脸图像超分辨率方法 |
| CN102880878B (zh) * | 2011-07-15 | 2015-05-06 | 富士通株式会社 | 一种基于单幅图像进行超分辨率分析的方法及系统 |
| CN102968775B (zh) * | 2012-11-02 | 2015-04-15 | 清华大学 | 基于超分辨率重建技术的低分辨率人脸图像的重建方法 |
| CN103150713B (zh) * | 2013-01-29 | 2015-12-09 | 南京理工大学 | 利用图像块分类稀疏表示与自适应聚合的图像超分辨方法 |
| CN104867111B (zh) * | 2015-03-27 | 2017-08-25 | 北京理工大学 | 一种基于分块模糊核集的非均匀视频盲去模糊方法 |
| CN105069765B (zh) * | 2015-07-22 | 2017-12-22 | 广东迅通科技股份有限公司 | 一种基于特征学习的模糊车牌重建方法 |
-
2017
- 2017-08-31 CN CN201710774351.1A patent/CN107563978A/zh active Pending
- 2017-12-19 WO PCT/CN2017/117166 patent/WO2019041660A1/zh unknown
- 2017-12-19 EP EP17923783.9A patent/EP3598385B1/en active Active
-
2019
- 2019-10-16 US US16/654,779 patent/US20200051228A1/en not_active Abandoned
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104484803A (zh) * | 2014-11-24 | 2015-04-01 | 苏州福丰科技有限公司 | 基于神经网络的三维人脸识别的手机支付方法 |
| CN105678252A (zh) * | 2016-01-05 | 2016-06-15 | 安阳师范学院 | 依据人脸三角网格自适应细分和高斯小波的迭代插值方法 |
| WO2017131343A1 (en) * | 2016-01-30 | 2017-08-03 | Samsung Electronics Co., Ltd. | A device for and method of enhancing quality of an image |
| CN107563978A (zh) * | 2017-08-31 | 2018-01-09 | 苏州科达科技股份有限公司 | 人脸去模糊方法及装置 |
Non-Patent Citations (1)
| Title |
|---|
| See also references of EP3598385A4 * |
Also Published As
| Publication number | Publication date |
|---|---|
| EP3598385A1 (en) | 2020-01-22 |
| CN107563978A (zh) | 2018-01-09 |
| EP3598385B1 (en) | 2021-06-16 |
| EP3598385A4 (en) | 2020-03-18 |
| US20200051228A1 (en) | 2020-02-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2019041660A1 (zh) | 人脸去模糊方法及装置 | |
| WO2022002150A1 (zh) | 一种视觉点云地图的构建方法、装置 | |
| WO2021174939A1 (zh) | 人脸图像的获取方法与系统 | |
| CN107481279B (zh) | 一种单目视频深度图计算方法 | |
| CN111243093B (zh) | 三维人脸网格的生成方法、装置、设备及存储介质 | |
| WO2022001236A1 (zh) | 三维模型生成方法、装置、计算机设备及存储介质 | |
| US20240046557A1 (en) | Method, device, and non-transitory computer-readable storage medium for reconstructing a three-dimensional model | |
| WO2018228436A1 (zh) | 双视角图像校准及图像处理方法、装置、存储介质和电子设备 | |
| CN113129425B (zh) | 一种人脸图像三维重建方法、存储介质及终端设备 | |
| CN109919971B (zh) | 图像处理方法、装置、电子设备及计算机可读存储介质 | |
| CN109685873B (zh) | 一种人脸重建方法、装置、设备和存储介质 | |
| CN109934847A (zh) | 弱纹理三维物体姿态估计的方法和装置 | |
| CN111080776B (zh) | 人体动作三维数据采集和复现的处理方法及系统 | |
| WO2022052782A1 (zh) | 图像的处理方法及相关设备 | |
| Liu et al. | On 3D face reconstruction via cascaded regression in shape space | |
| WO2022267939A1 (zh) | 图像处理方法及装置、计算机可读存储介质 | |
| CN116012432A (zh) | 立体全景图像的生成方法、装置和计算机设备 | |
| Firintepe et al. | The more, the merrier? A study on in-car IR-based head pose estimation | |
| CN116051736A (zh) | 一种三维重建方法、装置、边缘设备和存储介质 | |
| Han et al. | Inference-reconstruction variational autoencoder for light field image reconstruction | |
| WO2020151078A1 (zh) | 一种三维重建的方法和装置 | |
| Pini et al. | Learning to generate facial depth maps | |
| CN113610969A (zh) | 一种三维人体模型生成方法、装置、电子设备及存储介质 | |
| CN112270748A (zh) | 基于图像的三维重建方法及装置 | |
| Halperin et al. | Clear Skies Ahead: Towards Real‐Time Automatic Sky Replacement in Video |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 17923783 Country of ref document: EP Kind code of ref document: A1 |
|
| ENP | Entry into the national phase |
Ref document number: 2017923783 Country of ref document: EP Effective date: 20191017 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |