CN102663361A - Face image reversible geometric normalization method facing overall characteristics analysis - Google Patents
Face image reversible geometric normalization method facing overall characteristics analysis Download PDFInfo
- Publication number
- CN102663361A CN102663361A CN2012100965956A CN201210096595A CN102663361A CN 102663361 A CN102663361 A CN 102663361A CN 2012100965956 A CN2012100965956 A CN 2012100965956A CN 201210096595 A CN201210096595 A CN 201210096595A CN 102663361 A CN102663361 A CN 102663361A
- Authority
- CN
- China
- Prior art keywords
- mrow
- image
- msub
- mouth
- mtd
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000010606 normalization Methods 0.000 title claims abstract description 73
- 238000004458 analytical method Methods 0.000 title claims abstract description 15
- 230000002441 reversible effect Effects 0.000 title claims abstract description 12
- 230000009466 transformation Effects 0.000 claims abstract description 74
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims abstract description 51
- 238000000034 method Methods 0.000 claims abstract description 33
- 210000001508 eye Anatomy 0.000 claims abstract description 31
- 230000004927 fusion Effects 0.000 claims abstract description 25
- 238000013519 translation Methods 0.000 claims description 8
- 239000003550 marker Substances 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 5
- 230000005540 biological transmission Effects 0.000 claims description 3
- 230000003287 optical effect Effects 0.000 claims description 3
- YBJHBAHKTGYVGT-ZKWXMUAHSA-N (+)-Biotin Chemical compound N1C(=O)N[C@@H]2[C@H](CCCCC(=O)O)SC[C@@H]21 YBJHBAHKTGYVGT-ZKWXMUAHSA-N 0.000 claims description 2
- FEPMHVLSLDOMQC-UHFFFAOYSA-N virginiamycin-S1 Natural products CC1OC(=O)C(C=2C=CC=CC=2)NC(=O)C2CC(=O)CCN2C(=O)C(CC=2C=CC=CC=2)N(C)C(=O)C2CCCN2C(=O)C(CC)NC(=O)C1NC(=O)C1=NC=CC=C1O FEPMHVLSLDOMQC-UHFFFAOYSA-N 0.000 claims description 2
- 238000012545 processing Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 210000000887 face Anatomy 0.000 description 3
- 230000001815 facial effect Effects 0.000 description 3
- 238000003384 imaging method Methods 0.000 description 3
- 238000000513 principal component analysis Methods 0.000 description 3
- 230000000007 visual effect Effects 0.000 description 2
- 230000003321 amplification Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003199 nucleic acid amplification method Methods 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 210000000697 sensory organ Anatomy 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a face image reversible geometric normalization method facing overall characteristics analysis. A method based on secondary affine transformation is adopted. Firstly global affine is carried out which means that global rotation, zooming and cutting are carried out on a face image according to a standard structure definition of a face, such that in an input face image, coordinates of two eyes are at a same horizontal line, a two eye center distance is subjected to normalization according to a parameter d of a standard structure, and transverse and longitudinal proportions of the face are consistent with the standard structure. The image is cut into a magnitude of 2d*2d. Then geometric position normalization is carried out on a mouth area through a local affine transformation needle. Finally by utilizing image edge fusion, a standard geometric normalization face image is formed. An inverse normalization process is that according to a transformation parameter, a standard structure face image is recovered to a geometric structure characteristic before geometric normalization through secondary affine inverse transformation such that personalized topology structures of different face characteristic points are maintained. The face image which goes through geometric normalization can be used for face analysis based on an overall characteristic, such as super resolution reconstruction of the face image, face recognition, and the like.
Description
Technical Field
The invention relates to a geometric normalization method for a face image, in particular to a reversible geometric normalization method for the face image based on secondary affine transformation, which belongs to the field of face image processing.
Background
With the development of computer and multimedia technologies and the continuous progress of society, various fields such as security monitoring, visual communication, medical imaging, satellite remote sensing, electronic entertainment, computer vision and the like put forward wide demands on high-quality images and videos, and the high-quality images/videos can provide richer information and more real visual feelings, so that the method is the basis of many practical applications. The spatial resolution refers to the amount of information stored in an image, and is an important index for measuring the detail expressive power of the image. The real world continuous scene itself has rich information, but is affected by various factors such as imaging equipment, imaging environment, noise and the like, and the digital image acquired by the imaging equipment is generally a low-resolution image, which is difficult to meet the continuously improved practical application requirements. Especially images/videos from various handheld devices, vehicle-mounted video capture devices, wireless video sensor networks, and video monitoring devices in harsh natural conditions, etc., are often unsatisfactory in terms of resolution. Therefore, the Super-Resolution (SR) technology has become a research focus in the image processing field in recent years.
The human face is an important object in image processing and analysis, and has attracted wide attention for super-resolution restoration of human face images. Aiming at the face image, the method based on the integral face feature analysis obtains a good experimental result. Such methods perform feature analysis on the face image as a whole. A sample library of pairs of high resolution face images and corresponding low resolution face images is subjected to pca (principal Component analysis) or ica (identity Component analysis) analysis. The whole human face is decomposed into linear combinations of different characteristic faces by PCA. The sample library is expressed as a set of eigenfaces after PCA decomposition is carried out on a group of low-resolution eigenfaces and corresponding high-resolution eigenfaces. The super-resolution restoration is to decompose the low-resolution face into linear combinations of a plurality of low-resolution eigenfaces to obtain combination coefficients. The set of coefficients is then applied directly to the corresponding high resolution intrinsic face, resulting in a new high resolution face.
The method can obtain a result with higher resolution, but the good and bad face effect of the reconstruction result depends on whether the image in the sample library is well geometrically normalized or not. In reality, the sizes of human faces are different, and the positions of five sense organs are different from person to person, so that the geometric normalization of the human face image facing the overall features plays a key role, and is an important factor for determining the image result after super-resolution restoration. Existing methods are typically implemented by normalizing the position of both eyes. I.e. the coordinate position of the human eye is fixed by global rotation, scaling and cropping of the image of the human face. Doing so still leaves significant artifacts after restoration of misaligned nose and mouth, and the problem is not fundamentally solved. Meanwhile, the methods generally do not consider the change of the geometric normalization to the human face personalized proportion characteristic. The original human face personalized scale characteristics are destroyed after the input image is modified.
Disclosure of Invention
The invention aims to realize reversible geometric normalization of a human face image by a method based on secondary affine transformation, and respectively adopts three methods of global affine transformation, local affine transformation and image edge fusion according to the definition of a human face standard geometric structure.
The invention is realized by adopting the following technical means:
a human face geometric normalization method based on secondary affine transformation comprises the steps that a video acquisition device acquires a human face image, an optical signal of a target image is converted into a digital image signal, and the digital image signal is stored in a storage device of the acquisition device; reading in images through the existing USB and infrared interfaces by a computer, and carrying out geometric normalization of the human face based on secondary affine transformation in a processor; the normalization result can be stored in a database form and directly stored in a local hard disk, or stored in a network storage device or stored in a network; the normalization method mainly comprises the steps of face standard geometric structure definition, global affine transformation, local affine transformation and image edge fusion.
In the global affine transformation step, after the face image is marked according to the standard structure definition (figure 1) of the face, global rotation, scaling and cutting are carried out, so that the input face image is normalized according to the parameters d of the standard structure when the coordinates of two eyes are positioned on the same horizontal line and the center distance of the two eyes, and the transverse and longitudinal proportion of the face is consistent with the standard structure;
in the local affine transformation step, geometric position normalization is performed for the mouth region. The method mainly comprises the steps of mouth region division, local rotation and scaling and a converted mouth region;
in the image edge fusion step, edge fusion is carried out on the global transformation image and the local transformation image so as to form a standard geometric normalization human face image;
the image normalization method comprises the following steps:
reading in the collected face image from the database;
describing the face image by 6 feature point coordinates and relative positions thereof, and selecting P1～P6(the corresponding coordinate is (x)1，y1)～(x6，y6) Respectively corresponding to the coordinates of the left eye center, the right eye center, the nose tip, the left mouth corner, the mouth center and the right mouth corner;
calculating the relative position relationship F ═ d, r between the characteristic points1，r2，r3). Wherein,
i.e. d is the Euclidean distance between the centers of the eyes, r1The ratio of the vertical distance from the line of the eyes to the nose tip to d; r is2Is the ratio of the distance from the tip of the nose to the center point of the mouth to d, r3Is the ratio of the distance from the center point of the mouth to the corner of the mouth to d. The length and width of the face image are defined as 2d, and the edge of the image with the nearest distance between the centers of the eyes is 0.5 d;
marking characteristic points of a face image sample library and solving statistical average to obtain standard structure parameters F aiming at different human face images;
carrying out global affine transformation on the image defined by the standard structure of the face, and carrying out global rotation, scaling and cutting to enable the input face image to be in the same horizontal line with the coordinates of two eyes, and the central distance of the two eyes to be normalized according to the parameter d of the standard structure, so that the transverse and longitudinal proportions of the face are consistent with the standard structure;
and carrying out local affine transformation and carrying out geometric position normalization on the mouth region. The method mainly comprises the steps of mouth area division, local rotation and scaling, and edge fusion of the mouth area after transformation and an original image.
The edges of the mouth local image and the global image are fused to complete the geometric normalization of the face image;
storing the face image after geometric normalization into a database, and carrying out operations such as local storage or remote transmission through network storage equipment;
the inverse normalization process is opposite to the geometric normalization process, and starts from the local inverse affine and then reaches the global inverse affine;
the foregoing global affine transformation step is as follows:
for the input face image, artificial characteristic points (P) are carried out1～P6) Notation, according to P1And P2Coordinate calculation of global image rotation angle, as in equation (5)) Shown in the figure:
(5)
wherein alpha is P1And P2The angle of the line to the horizontal. The image is rotated by an angle theta-alpha. The coordinates before and after rotation are expressed by the following formula (6):
(6)
defining the upper left corner of the rotated image as a new coordinate system origin, translating the coordinate origin relative to the original image origin, and setting the width and height of the original image as W and H respectively, wherein the coordinate transformation relation of the image rotating counterclockwise is shown as formula (7):
(7)
where (x, y) is the original image coordinates and (x ', y') is the rotated image coordinates. Similarly, the coordinate relationship between the clockwise rotation image and the original image is:
(8)
the feature points of the two eyes of the rotated image are located on the same horizontal line. And calculating the horizontal and vertical scaling coefficients, and cutting to make the coordinates of the eyes and the nose tip consistent with the standard geometric structure of the human face. The transverse and longitudinal scaling factors are respectively shown in formulas (9) and (10):
Sh＝(x′2-x′1)/d (9)
Sy＝(y′3-y′1)/r1d
(10)
wherein, the coordinate point x'1、x′2、y′1And y'3As coordinates of the rotated feature points, r1And d is a standard structural parameter defined by formula (2), i.e., d is the Euclidean distance between the centers of the eyes, r1Is the ratio of the perpendicular distance from the line of the eyes to the tip of the nose to d. To ensure the reversibility of the normalization, the parameters α and abscissaThe ratio S to the vertical scaling factor Sh/Sv is saved as the inverse transform parameter.
The specific steps of the local affine transformation aiming at the geometric normalization of the mouth region are as follows:
a rectangular mouth area is defined, and the length and the height of the rectangle are respectively shown as formulas (11) and (12):
w＝a1(x6-x4) (11)
h＝a2(y5-y3) (12)
the coordinates of the vertex at the upper left corner of the rectangle are shown in equation (13):
(xo，yo)＝(x5-w/2，(y5+y3)/2) (13)
the triangular areas of the left lower corner and the right lower corner of the rectangle are excluded from the mouth area. Two right-angle sides of the triangular area are w/4 and h/4 respectively. Selection of a by testing the CAS-PEAL-R1 database1＝1.7，a2＝1.1；
Calculating the local rotation angle of the mouth area through the coordinates of the characteristic points P4-P6 as shown in formula (14):
the local scaling factor is calculated as shown in equation (15):
Sm＝Sp＝Sq＝(2r3d)/(x′6-x′4) (15)
wherein, x'4And x'6Is a characteristic point P4And P6And transverse coordinates in the partial image after the rotation transformation. Local scaling of the mouth takes lateral and longitudinal equal scaling;
the rotated and scaled mouth image is translated such that the transformed P5Coordinate and P in human face standard structure5Coordinate coincidence, translation offset is noted as (Δ)x，Δy)；
All the parameters F for the inverse transformation are obtainedinv＝[α，S，β，Sm，Δx，Δy]；
The edge fusion step for the global transformation image and the local transformation image is as follows:
and establishing a Label image Label. The marker image size is 2d × 2 d;
and marking each pixel point of the image by a numerical value of 0-3. Initializing all mark values of the image to 0; after the mouth area is defined, pixels at corresponding positions of the mouth area are located, and the value of the marked image is 1; after local rotation, scaling and translation, the new mouth region covers a region marker image marker value of 2. The specific meaning of such a marked image is: the position of the value 0 is a pixel area which is not involved in local transformation; the area with the value of 1 belongs to the original mouth area and is not covered by the new mouth area; the pixel with the value of 2 is the position of the new mouth area which is not overlapped with the original mouth area; the pixel position with the value of 3 is the position where the new mouth area is superposed with the original mouth area;
for the pixels in the 0 area, directly adopting the pixels in the global image as final normalized image pixels;
for the pixels in the area 2, the new mouth area covers the original pixel points after local affine transformation, so the pixels in the new mouth area directly replace the covered original pixel points;
the area 3 is the superposition part of the new mouth area and the original image mouth area, and a fusion result is obtained by adopting a weighted average method so as to avoid obvious edge artifacts in the image; the weight is defined as follows: p transformed by local affine5(x5，y5) Constructing a two-dimensional Gaussian function for the center is shown in equation (16):
(16)
wherein i ═ x-x5，j＝y-y5All are image coordinates in the local image;
wm＝w×Smand hm＝h×SmRespectively the width and the height of the local affine transformation rear mouth part image;
the synthesized pixel of the overlap region (i, j) is represented by formula (17):
p(i，j)＝Weight(i，j)pG+(1-Weight(i，j))pL.
(17)
PGis the pixel value, P, of a pixel point in the global image overlap regionLPixel values of pixel points overlapped with the global image in the mouth region;
the area 1 pixels are areas which belong to the original mouth area but are not covered by the new mouth area, and the pixels in the area 1 in the fused image are filled through one-dimensional linear interpolation;
traversing the Label graph, if the current mark is a 1 area, counting the lengths RunWidth and RunHeight of the area which is continuously and transversely 1 with the current point as the starting point;
selecting a direction with a smaller continuous length to ensure that the interpolation pixel points have good continuity, and performing linear interpolation according to the pixel values of the 0 regions at the two ends of the 1 region in the fusion image to obtain the fusion pixel value of the 1 region;
if RunHeight is less than RunWidth, fusing pixel points:
(18)
wherein QG(i1J) is the uppermost first pixel not belonging to region 1 of the same row of interpolation pixels, QL(i2J) is the first pixel point which is not in the 1 area at the lowest end of the same row of the interpolation pixel points.
If RunWidth is less than RunHeight, fusing pixel points:
(19)
wherein QM(i，j1) The first pixel point not belonging to the 1 region at the leftmost end of the same line of interpolation pixel points, QN(i，j2) The first pixel point which is not in the region 1 at the rightmost end of the same line of the interpolation pixel points.
Modifying the Label 1 in the Label image corresponding to the interpolated pixel into 0, and providing help for the linear interpolation of the unfused 1-region pixel;
compared with the prior art, the invention has the following obvious advantages and beneficial effects:
the face image is defined according to the standard structure of the face, global affine transformation is carried out, and the face image is rotated, scaled and cut, so that the input face image is in the same horizontal line with the coordinates of two eyes, the center distance of the two eyes is normalized according to the parameter d of the standard structure, and the transverse and longitudinal proportion of the face is consistent with the standard structure. And then applying local affine transformation, and carrying out geometric position normalization on the mouth region, wherein the geometric position normalization comprises mouth region division, local rotation and scaling and the transformed mouth region. Finally, carrying out image edge fusion, and carrying out edge fusion on the global transformation image and the local transformation image to form a complete geometric normalization face image;
the method fully utilizes the characteristic of facial image facial feature position approximation, and uniformly performs reversible facial image geometric normalization based on secondary affine transformation on the images in the database before performing super-resolution restoration based on global feature PCA, so that the super-resolution restoration result is well improved, and meanwhile, the original features of the images are maintained by the reversible normalization method.
The invention has the characteristics that:
1. the 6-point coordinate is adopted to describe the topological structure of the main features of the face, so that the method is simple and clear; a standard face structure method based on statistics to determine 6-point description is provided;
2. the global affine transformation and the local affine transformation are utilized to respectively carry out geometric normalization on the global structure and the local structure of the mouth of the human face, the principle is simple, and the implementation is easy;
3. based on the marked edge region fusion technology, image artifacts introduced by local structure adjustment can be effectively removed;
4. the human face geometric normalization process is parameterized, and the inverse transformation of the normalization process can be realized. The problem that the change of the normalization processing on the face structure influences the face processing result is solved.
Description of the drawings:
FIG. 1, Standard Structure definition of human faces;
FIG. 2, (a) coordinate mapping after counterclockwise rotation, scaling and clipping; (b) coordinate corresponding relation after clockwise rotation;
FIG. 3, (a) a mouth region division diagram; (b) partially transforming the labeled graph;
FIG. 4, (a) a diagram of generation of a training sample library; (b) the invention relates to a human face super-resolution restoration block diagram.
Fig. 5(a) is a face image reconstruction result without normalization processing (b) is a face image reconstruction result after normalization processing.
The specific implementation mode is as follows:
the embodiments of the invention are described in detail below with reference to the accompanying drawings:
the method comprises the steps that a video acquisition device acquires a face image, converts an optical signal of a target image into a digital image signal and stores the digital image signal in a storage of the acquisition device; reading in images through the existing USB and infrared interfaces by a computer, and carrying out geometric normalization of the human face based on secondary affine transformation in a processor; the normalization result can be stored in a database form and directly stored in a local hard disk, or stored in a network storage device or stored in a network; the normalization method mainly comprises the steps of face standard geometric structure definition, global affine transformation, local affine transformation and image edge fusion.
Definition of standard geometric structure of human face
As shown in figure 1, 6 characteristic point coordinates and relative positions thereof are used for describing a face image, and P is selected1～P6(the corresponding coordinate is (x)1，y1)～(x6，y6) Respectively corresponding to the coordinates of the left eye center, the right eye center, the nose tip, the left mouth corner, the mouth center and the right mouth corner;
calculating the relative position relationship F ═ d, r between the characteristic points1，r2，r3). Wherein,
i.e. d is the Euclidean distance between the centers of the eyes, r1The ratio of the vertical distance from the line of the eyes to the nose tip to d; r is2Is the ratio of the distance from the tip of the nose to the center point of the mouth to d, r3Is the ratio of the distance from the center point of the mouth to the corner of the mouth to d. The length and width of the face image are defined as 2d, and the edge of the image with the nearest distance between the centers of the eyes is 0.5 d;
marking characteristic points of a face image sample library and solving statistical average to obtain standard structure parameters F aiming at different human face images;
(II) Global affine transformation
The global affine transformation step is as follows:
for the input face image, artificial characteristic points (P) are carried out1～P6) Notation, according to P1And P2And (3) calculating the global image rotation angle by coordinates, wherein the global image rotation angle is expressed by formula (5):
(5)
wherein alpha is P1And P2The angle of the line to the horizontal. The image is rotated by an angle theta-alpha. The coordinates before and after rotation are expressed by the following formula (6):
(6)
defining the upper left corner of the rotated image as a new coordinate system origin, then the origin of coordinates is translated with respect to the origin of the original image, and if the width and height of the original image are W and H, respectively, as shown in fig. 2, then the coordinate transformation relation of the image rotating counterclockwise is as shown in equation (7):
where (x, y) is the original image coordinates and (x ', y') is the rotated image coordinates. Similarly, the coordinate relationship between the clockwise rotation image and the original image is:
(8)
the feature points of the two eyes of the rotated image are located on the same horizontal line. And calculating the horizontal and vertical scaling coefficients, and cutting to make the coordinates of the eyes and the nose tip consistent with the standard geometric structure of the human face. The transverse and longitudinal scaling factors are respectively shown in formulas (9) and (10):
Sh＝(x′2-x′1)/d (9)
Sy＝(y′3-y′1)/r1d
(10)
wherein, the coordinate point x'1、x′2、y′1And y'3As coordinates of the rotated feature points, r1And d is the criterion defined in section 2.1 in order to ensure the reversibility of the normalization, the parameter α and the ratio of the horizontal to vertical scaling factor S Sh/Sv are stored as inverse transform parameters. Meanwhile, according to the parameters of rotation, scaling and clipping, the face characteristic point P is updated1～P6The coordinates of (a);
(III) local affine transformations
Local affine transformation, aiming at the geometric normalization of a mouth region, the specific steps are as follows:
the rectangular mouth area is defined as shown in fig. 3(a), and the length and height of the rectangle are respectively shown in formulas (11) and (12):
w＝a1(x6-x4) (11)
h＝a2(y5-y3) (12)
the coordinates of the vertex at the upper left corner of the rectangle are shown in equation (13):
(xo，yo)＝(x5-w/2，(y5+y3)/2) (13)
the triangular areas of the left lower corner and the right lower corner of the rectangle are excluded from the mouth area. The two right-angle sides of the triangular area are l4 and h/4 respectively. Selection of a by testing the CAS-PEAL-R1 database1＝1.7，a2＝1.1；
Calculating the local rotation angle of the mouth area through the coordinates of the characteristic points P4-P6 as shown in formula (14):
the local scaling factor is calculated as shown in equation (15):
Sm＝Sp＝Sq＝(2r3d)/(x′6-x′4) (15)
x′4and x'6Is a characteristic point P4And P6And transverse coordinates in the partial image after the rotation transformation. Local scaling of the mouthScaling to and longitudinally;
the rotated and scaled mouth image is translated such that the transformed P5Coordinate and P in human face standard structure5Coordinate coincidence, translation offset is noted as (Δ)x，Δy)；
All the parameters F for the inverse transformation are obtainedinv＝[α，S，β，Sm，Δx，Δy]；
(IV) image edge blending
The edge fusion step for the global transformation image and the local transformation image is as follows:
and establishing a Label image Label. The marker image size is 2d × 2 d;
and marking each pixel point of the image by a numerical value of 0-3. All flag values are initialized to 0; after the mouth area is defined, pixels at corresponding positions of the mouth area are located, and the value of the marked image is 1; after local rotation, scaling and translation, the area covered by the new mouth area and the corresponding marker image marker value is 2. The specific meaning of such a marked image is: the position of the value 0 is a pixel area which is not involved in local transformation; the area with the value of 1 and the pixel which belongs to the original mouth area and is not covered by the new mouth area; the pixel with the value of 2 is the position of the new mouth area which is not overlapped with the original mouth area; the pixel position with the area value of 3 is the position where the new mouth area is superposed with the original mouth area;
for the pixels in the 0 area, directly adopting the pixels in the global image as final normalized image pixels;
for the pixels in the area 2, the new mouth area covers the original pixel points after local affine transformation, so the pixels in the new mouth area directly replace the original pixel points covered by the new mouth area;
the area 3 is the superposition part of the new mouth area and the original image mouth area, and a fusion result is obtained by adopting a weighted average method so as to avoid obvious edge artifacts in the image; the weight is defined as follows: by local affineTransformed P5(x5，y5) Constructing a two-dimensional Gaussian function for the center is shown in equation (16):
(16)
wherein i ═ x-x5，j＝y-y5All are image coordinates in the local image;
wm＝w×Smand hm＝h×SmRespectively the width and the height of the local affine transformation rear mouth part image;
the synthesized pixel of the overlap region (i, j) is represented by formula (17):
p(i，j)＝Weight(i，j)PG+(1-Weight(i，j))PL.
(17)
PGis the pixel value, P, of a pixel point in the global image overlap regionLPixel values of pixel points overlapped with the global image in the mouth region;
the area 1 pixels are areas which belong to the original mouth area but are not covered by the new mouth area, and the pixels in the area 1 in the fused image are filled through one-dimensional linear interpolation;
traversing the Label graph, if the current mark is a 1 area, counting the lengths RunWidth and RunHeight of the area which is continuously and transversely 1 with the current point as the starting point;
selecting a direction with a smaller continuous length to ensure that the interpolation pixel points have good continuity, and performing linear interpolation according to the pixel values of the 0 regions at the two ends of the 1 region in the fusion image to obtain the fusion pixel value of the 1 region;
if RunHeight is less than RunWidth, fusing pixel points:
(18)
wherein QG(i1J) is the uppermost first pixel not belonging to region 1 of the same row of interpolation pixels, QL(i2J) is the first pixel point which is not in the 1 area at the lowest end of the same row of the interpolation pixel points.
If RunWidth is less than RunHeight, fusing pixel points:
(19)
wherein QM(i，j1) The first pixel point not belonging to the 1 region at the leftmost end of the same line of interpolation pixel points, QN(i，j2) The first pixel point which is not in the region 1 at the rightmost end of the same line of the interpolation pixel points.
Modifying the Label 1 in the Label image corresponding to the interpolated pixel into 0, and providing help for the linear interpolation of the unfused 1-region pixel;
storing the face image after geometric normalization into a database, and carrying out operations such as local storage or remote transmission through network storage equipment; the normalized face image library can be used for PCA analysis, establishing a sample library of face image analysis, and being used for face recognition or face super-resolution reconstruction and the like.
(V) inverse normalization
And according to the standard structure definition of the human face, a local mouth area is defined. According to FinvLocal translation parameter (Δ) among the parametersx，Δy) Local scaling factor SmAnd the local rotation angle beta performs translation, scaling and rotation of the mouth image;
and fusing the edge of the image subjected to local affine transformation and the edge of the global image.
And scaling and rotating the fused image according to the S and the alpha in the reversible parameters to obtain the human face image after inverse normalization. And S is the ratio of the horizontal scaling coefficient and the vertical scaling coefficient of the global affine transformation in the normalization. In the inverse normalization, the height of the target image can be calculated according to the S and the width of the target image, so that inverse scaling of different proportions in the transverse direction and the longitudinal direction is realized.
And performing integral rotation of the image according to the global angle parameter alpha.
Fig. 4 illustrates a block diagram of a system for applying the method of the present invention to super-resolution reconstruction of face images. Fig. 4(a) shows a process of generating a sample of a face image library. The high-resolution face image which is geometrically normalized by the inventionThe library is downsampled to generate a corresponding low resolution image. And carrying out PCA analysis on the high-resolution face library and the low-resolution face library respectively to obtain corresponding feature vector groups serving as sample libraries for face image super-resolution reconstruction. Fig. 4(b) is a super-resolution reconstruction process based on global feature analysis for a face image. Firstly, the geometric normalization based on the invention is carried out on the low-resolution face image to obtain the inverse transformation parameter Finv(ii) a And then, performing linear representation on the normalized face image by utilizing the PCA feature vector group of the low-resolution images in the sample library to obtain a linear combination coefficient C. And carrying out linear combination on the combination coefficient C and the PCA characteristic vector group of the high-resolution face to obtain an initial estimation image of the high-resolution face image. F-based initial resultsinvAnd performing inverse normalization processing on the parameters to obtain a high-resolution face image reconstruction result. FIG. 5 illustrates that the method of the present invention is applied to the super-resolution reconstruction process of the face image, which can effectively improve the quality of the reconstructed image. Wherein, fig. 5(a) is the face image super-resolution reconstruction result without normalization processing, and fig. 5(b) is the face image super-resolution reconstruction result after normalization processing. The first attached table shows the average PSNR values of the test samples with twice amplification results after normalization and non-normalization processing are performed on 300 sample images respectively in the image objective quality evaluation. It can be seen from the table that the results obtained for the normalized samples are significantly better than the set of samples that were not normalized. In the face image super-resolution reconstruction scheme, the method can retain the personalized geometric characteristics of the original face image and improve the effectiveness of the whole face analysis.
TABLE-CAS-PEAL-R1 database image normalization test PSNR mean value (2 Xmagnification)
| PSNR(db) | |
| Normalized result | 29.026 |
| Results without normalization | 27.421 |
Claims (4)
1. A face image reversible geometric normalization method facing integral feature analysis comprises the steps that a video acquisition device acquires a face image, an optical signal of a target image is converted into a digital image signal, and the digital image signal is stored in a storage device of the acquisition device; reading in images through the existing USB and infrared interfaces by a computer, and carrying out geometric normalization of the human face based on secondary affine transformation in a processor; the normalization result can be stored in a database form and directly stored in a local hard disk, or stored in a network storage device or stored in a network; the normalization method mainly comprises the steps of face standard geometric structure definition, global affine transformation, local affine transformation and image edge fusion;
in the global affine transformation step, after a face image is defined according to a standard structure of a face, global rotation, scaling and cutting are carried out, so that the input face image is enabled to have the two-eye coordinates in the same horizontal line, the center distance of the two eyes is normalized according to a parameter d of the standard structure, and the transverse and longitudinal proportions of the face are consistent with the standard structure;
in the local affine transformation step, geometric position normalization is carried out on a mouth region, and the main steps include mouth region division, local rotation and scaling and the mouth region after transformation;
in the image edge fusion step, edge fusion is carried out on the global transformation image and the local transformation image so as to form a standard geometric normalization human face image;
the image normalization algorithm comprises the following steps:
reading in the collected face image from the database;
describing the face image by 6 feature point coordinates and relative positions thereof, and selecting P1～P6(the corresponding coordinate is (x)1，y1)～(x6，y6) Respectively corresponding to the coordinates of the left eye center, the right eye center, the nose tip, the left mouth corner, the mouth center and the right mouth corner;
calculating the relative position relationship F ═ d, r between the characteristic points1，r2，r3) (ii) a Wherein,
i.e. d is the Euclidean distance between the centers of the eyes, r1The ratio of the vertical distance from the line of the eyes to the nose tip to d; r is2Is the ratio of the distance from the tip of the nose to the center point of the mouth to d, r3The ratio of the distance from the center point of the mouth to the angle of the mouth to d; the length and width of the face image are defined as 2d, and the edge of the image with the nearest distance between the centers of the eyes is 0.5 d;
marking characteristic points of a face image sample library and solving statistical average to obtain standard structure parameters F aiming at different human face images;
carrying out global affine transformation on the image defined by the standard structure of the face, and carrying out global rotation, scaling and cutting to enable the input face image to be in the same horizontal line with the coordinates of two eyes, and the central distance of the two eyes to be normalized according to the parameter d of the standard structure, so that the transverse and longitudinal proportions of the face are consistent with the standard structure;
carrying out local affine transformation and carrying out geometric position normalization on the mouth region; dividing a mouth region, locally rotating and scaling, and fusing the transformed mouth region and the edge of an original image;
the edges of the mouth local image and the global image are fused to complete the geometric normalization of the face image;
storing the face image after geometric normalization into a database, and carrying out operations such as local storage or remote transmission through network storage equipment;
the inverse normalization process is the inverse of the geometric normalization process described above, starting with the local inverse affine and going to the global inverse affine.
2. The reversible geometric normalization method for human face images oriented to integral feature analysis according to claim 1, characterized in that: the global affine transformation comprises the following specific steps:
2.1: for the input face image, artificial characteristic points (P) are carried out1～P6) Notation, according to P1And P2And (3) calculating the global image rotation angle by coordinates, wherein the global image rotation angle is expressed by formula (5):
2.2: wherein alpha is P1And P2Angle of the connecting line and the horizontal line; rotating the image by an angle theta-alpha; the coordinates before and after rotation are expressed by the following formula (6):
2.3: defining the upper left corner of the rotated image as a new coordinate system origin, translating the coordinate origin relative to the original image origin, and setting the width and height of the original image as W and H respectively, wherein the coordinate transformation relation of the image rotating counterclockwise is shown as formula (7):
2.4: wherein (x, y) is the original image coordinate, and (x ', y') is the rotated image coordinate; similarly, the coordinate relationship between the clockwise rotation image and the original image is:
2.5: the feature points of the two eyes of the rotated image are positioned on the same horizontal line; calculating the horizontal and vertical scaling coefficients, and cutting to make the coordinates of the eyes and the nose tip consistent with the standard geometric structure of the human face; the transverse and longitudinal scaling factors are respectively shown in formulas (9) and (10):
Sh＝(x′2-x′1)/d (9)
Sy＝(y′3-y′1)/r1d
(10)
2.6: wherein, the coordinate point x'1、x′2、y′1And y'3Is the feature point coordinate after rotation, d is the Euclidean distance between the centers of the two eyes, r1The ratio of the vertical distance from the line of the eyes to the tip of the nose to d, the parameter α and the ratio of the horizontal to vertical scaling factor S Sh/Sv are stored as inverse transform parameters.
3. The reversible geometric normalization method for human face images oriented to integral feature analysis according to claim 1, characterized in that: the local affine transformation specifically comprises the following steps aiming at geometric normalization of a mouth region:
3.1: a rectangular mouth area is defined, and the length and the height of the rectangle are respectively shown as formulas (11) and (12):
w＝a1(x6-x4) (11)
h＝a2(y5-y3) (12)
the coordinates of the vertex at the upper left corner of the rectangle are shown in equation (13):
(xo，yo)＝(x5-w/2，(y5+y3)/2) (13)
excluding the triangular areas at the left lower corner and the right lower corner of the rectangle from the mouth area; two right-angle sides of the triangular area are w/4 and h/4 respectively; selection of a by testing the CAS-PEAL-R1 database1＝1.7，a2＝1.1；
3.2: calculating the local rotation angle of the mouth area through the coordinates of the characteristic points P4-P6 as shown in formula (14):
the local scaling factor is calculated as shown in equation (15):
Sm＝Sp＝Sq＝(2r3d)/(x′6-x′4) (15)
wherein, x'4And x'6Is a characteristic point P4And P6Transverse coordinates in the partial image after the rotation transformation; local scaling of the mouth takes lateral and longitudinal equal scaling;
the rotated and scaled mouth image is translated such that the transformed P5Coordinate and P in human face standard structure5Coordinate coincidence, translation offset is noted as (Δ)x，Δy)；
Obtaining all parameters F for inverse transformationinv＝[α，S，β，Sm，Δx，Δy]。
4. The reversible geometric normalization method for human face images oriented to integral feature analysis according to claim 1, characterized in that: the method is characterized in that the global transformation image and the local transformation image are subjected to edge fusion, and the method specifically comprises the following steps:
4.1: establishing a Label image Label; the marker image size is 2d × 2 d;
4.2: marking each pixel point of the image by a numerical value of 0-3; initializing all mark values of the image to 0; after the mouth area is defined, pixels at corresponding positions of the mouth area are located, and the value of the marked image is 1; after local rotation, zooming and translation, the area covered by the new mouth area and the corresponding mark value of the mark image are 2; the specific meaning of such a marked image is: the position with the value of 0 is a pixel area related to local transformation; the area with the value of 1 and the pixel which belongs to the original mouth area and is not covered by the new mouth area; the pixel with the value of 2 is the position of the new mouth area which is not overlapped with the original mouth area; the pixel position with the area value of 3 is the position where the new mouth area is superposed with the original mouth area;
for the pixels in the 0 area, directly adopting the pixels in the global image as final normalized image pixels;
for the pixels in the area 2, the new mouth area covers the original pixel points after local affine transformation, so the pixels in the new mouth area directly replace the covered original pixel points;
the area 3 is a superposition part of a new mouth area and a non-mouth area of the original image, and a fusion result is obtained by adopting a weighted average method so as to avoid obvious edge artifacts in the image; the weight is defined as follows: p transformed by local affine5(x5，y5) Constructing a two-dimensional Gaussian function for the center is shown in equation (16):
wherein i ═ x-x5，i＝y-y5All are image coordinates in the local image;
wm＝w×Smand hm＝h×SmRespectively the width and the height of the local affine transformation rear mouth part image;
the synthesized pixel of the overlap region (i, j) is represented by formula (17):
p(i，j)＝Weight(i，j)pG+(1-Weight(i，j))pL.
(17)
PGis the pixel value, p, of a pixel point in the global image overlap regionLPixel values of pixel points at corresponding positions of the mouth region;
4.3: the area 1 pixels are areas which belong to the original mouth area but are not covered by the new mouth area, and the pixels in the area 1 in the fused image are filled through interpolation;
traversing the Label graph, if the current mark is a 1 area, counting the lengths RunWidth and RunHeight of the area which is continuously and transversely 1 with the current point as the starting point;
selecting a direction with a smaller continuous length to ensure that the interpolation pixel points have good continuity, and performing linear interpolation according to the pixel values of the 0 regions at the two ends of the 1 region in the fusion image to obtain the fusion pixel value of the 1 region;
if RunHeight is less than RunWidth, fusing pixel points:
wherein QG(i1J) is the uppermost first pixel not belonging to region 1 of the same row of interpolation pixels, QL(i2J) is the first pixel point which is not in the 1 area at the lowest end of the same row of the interpolation pixel points;
if RunWidth is less than RunHeight, fusing pixel points:
wherein QM(i，j1) The first pixel point not belonging to the 1 region at the leftmost end of the same line of interpolation pixel points, QN(i，j2) The first pixel point which is not in the region 1 at the rightmost end of the same line of the interpolation pixel points;
and modifying the Label 1 in the Label image corresponding to the interpolated pixel into 0, and providing assistance for the linear interpolation of the unfused 1-region pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210096595.6A CN102663361B (en) | 2012-04-01 | 2012-04-01 | Face image reversible geometric normalization method facing overall characteristics analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210096595.6A CN102663361B (en) | 2012-04-01 | 2012-04-01 | Face image reversible geometric normalization method facing overall characteristics analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102663361A true CN102663361A (en) | 2012-09-12 |
| CN102663361B CN102663361B (en) | 2014-01-01 |
Family
ID=46772845
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210096595.6A Expired - Fee Related CN102663361B (en) | 2012-04-01 | 2012-04-01 | Face image reversible geometric normalization method facing overall characteristics analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102663361B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103475876A (en) * | 2013-08-27 | 2013-12-25 | 北京工业大学 | Learning-based low-bit-rate compression image super-resolution reconstruction method |
| CN103914676A (en) * | 2012-12-30 | 2014-07-09 | 杭州朗和科技有限公司 | Method and apparatus for use in face recognition |
| CN106778577A (en) * | 2016-12-06 | 2017-05-31 | 浙江水马环保科技有限公司 | Water purifier user's personal identification method |
| CN106778578A (en) * | 2016-12-06 | 2017-05-31 | 浙江水马环保科技有限公司 | Water purifier method for identifying ID |
| CN107182218A (en) * | 2015-12-31 | 2017-09-19 | 深圳先进技术研究院 | A kind of authentication method and device |
| CN107506697A (en) * | 2017-07-29 | 2017-12-22 | 广东欧珀移动通信有限公司 | Anti-fake processing method and related product |
| CN107945102A (en) * | 2017-10-23 | 2018-04-20 | 深圳市朗形网络科技有限公司 | A kind of picture synthetic method and device |
| CN108492344A (en) * | 2018-03-30 | 2018-09-04 | 中国科学院半导体研究所 | A kind of portrait-cartoon generation method |
| CN108615014A (en) * | 2018-04-27 | 2018-10-02 | 京东方科技集团股份有限公司 | A kind of detection method of eye state, device, equipment and medium |
| CN108875556A (en) * | 2018-04-25 | 2018-11-23 | 北京旷视科技有限公司 | Method, apparatus, system and the computer storage medium veritified for the testimony of a witness |
| CN109855739A (en) * | 2019-01-04 | 2019-06-07 | 三峡大学 | Power equipment infrared measurement of temperature method and device based on affine transformation |
| CN109948555A (en) * | 2019-03-21 | 2019-06-28 | 于建岗 | Human face super-resolution recognition methods based on video flowing |
| CN111028251A (en) * | 2019-12-27 | 2020-04-17 | 四川大学 | Dental picture cutting method, system, equipment and storage medium |
| CN111028354A (en) * | 2018-10-10 | 2020-04-17 | 成都理工大学 | Image sequence-based model deformation human face three-dimensional reconstruction scheme |
| CN111259711A (en) * | 2018-12-03 | 2020-06-09 | 北京嘀嘀无限科技发展有限公司 | Lip movement identification method and system |
| CN112184593A (en) * | 2020-10-14 | 2021-01-05 | 北京字跳网络技术有限公司 | Key point determination method, device, equipment and computer readable medium |
| CN112308043A (en) * | 2020-11-26 | 2021-02-02 | 腾讯科技（深圳）有限公司 | Image processing method, image processing apparatus, and computer-readable storage medium |
| CN112464699A (en) * | 2019-09-06 | 2021-03-09 | 富士通株式会社 | Image normalization method, system and readable medium for face analysis |
| CN113643187A (en) * | 2020-04-27 | 2021-11-12 | 京东方科技集团股份有限公司 | Image restoration method and device |
| CN114972008A (en) * | 2021-11-04 | 2022-08-30 | 华为技术有限公司 | Coordinate restoration method and device and related equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1786980A (en) * | 2005-12-08 | 2006-06-14 | 上海交通大学 | Melthod for realizing searching new position of person's face feature point by tow-dimensional profile |
| US7778490B2 (en) * | 2003-01-13 | 2010-08-17 | Koninklijke Philips Electronics N.V. | Method of image registration and medical image data processing apparatus |
-
2012
- 2012-04-01 CN CN201210096595.6A patent/CN102663361B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7778490B2 (en) * | 2003-01-13 | 2010-08-17 | Koninklijke Philips Electronics N.V. | Method of image registration and medical image data processing apparatus |
| CN1786980A (en) * | 2005-12-08 | 2006-06-14 | 上海交通大学 | Melthod for realizing searching new position of person's face feature point by tow-dimensional profile |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103914676B (en) * | 2012-12-30 | 2017-08-25 | 杭州朗和科技有限公司 | A kind of method and apparatus used in recognition of face |
| CN103914676A (en) * | 2012-12-30 | 2014-07-09 | 杭州朗和科技有限公司 | Method and apparatus for use in face recognition |
| CN103475876B (en) * | 2013-08-27 | 2016-06-22 | 北京工业大学 | A kind of low bit rate compression image super-resolution rebuilding method based on study |
| CN103475876A (en) * | 2013-08-27 | 2013-12-25 | 北京工业大学 | Learning-based low-bit-rate compression image super-resolution reconstruction method |
| CN107182218A (en) * | 2015-12-31 | 2017-09-19 | 深圳先进技术研究院 | A kind of authentication method and device |
| CN106778578A (en) * | 2016-12-06 | 2017-05-31 | 浙江水马环保科技有限公司 | Water purifier method for identifying ID |
| CN106778577A (en) * | 2016-12-06 | 2017-05-31 | 浙江水马环保科技有限公司 | Water purifier user's personal identification method |
| CN107506697A (en) * | 2017-07-29 | 2017-12-22 | 广东欧珀移动通信有限公司 | Anti-fake processing method and related product |
| US11151398B2 (en) | 2017-07-29 | 2021-10-19 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Anti-counterfeiting processing method, electronic device, and non-transitory computer-readable storage medium |
| CN107945102A (en) * | 2017-10-23 | 2018-04-20 | 深圳市朗形网络科技有限公司 | A kind of picture synthetic method and device |
| CN108492344A (en) * | 2018-03-30 | 2018-09-04 | 中国科学院半导体研究所 | A kind of portrait-cartoon generation method |
| CN108875556A (en) * | 2018-04-25 | 2018-11-23 | 北京旷视科技有限公司 | Method, apparatus, system and the computer storage medium veritified for the testimony of a witness |
| CN108615014A (en) * | 2018-04-27 | 2018-10-02 | 京东方科技集团股份有限公司 | A kind of detection method of eye state, device, equipment and medium |
| CN108615014B (en) * | 2018-04-27 | 2022-06-21 | 京东方科技集团股份有限公司 | Eye state detection method, device, equipment and medium |
| CN111028354A (en) * | 2018-10-10 | 2020-04-17 | 成都理工大学 | Image sequence-based model deformation human face three-dimensional reconstruction scheme |
| CN111259711A (en) * | 2018-12-03 | 2020-06-09 | 北京嘀嘀无限科技发展有限公司 | Lip movement identification method and system |
| CN109855739A (en) * | 2019-01-04 | 2019-06-07 | 三峡大学 | Power equipment infrared measurement of temperature method and device based on affine transformation |
| CN109948555B (en) * | 2019-03-21 | 2020-11-06 | 于建岗 | Face super-resolution identification method based on video stream |
| CN109948555A (en) * | 2019-03-21 | 2019-06-28 | 于建岗 | Human face super-resolution recognition methods based on video flowing |
| CN112464699A (en) * | 2019-09-06 | 2021-03-09 | 富士通株式会社 | Image normalization method, system and readable medium for face analysis |
| CN111028251A (en) * | 2019-12-27 | 2020-04-17 | 四川大学 | Dental picture cutting method, system, equipment and storage medium |
| CN113643187A (en) * | 2020-04-27 | 2021-11-12 | 京东方科技集团股份有限公司 | Image restoration method and device |
| CN113643187B (en) * | 2020-04-27 | 2024-08-13 | 京东方科技集团股份有限公司 | Image restoration method and device |
| CN112184593A (en) * | 2020-10-14 | 2021-01-05 | 北京字跳网络技术有限公司 | Key point determination method, device, equipment and computer readable medium |
| CN112308043A (en) * | 2020-11-26 | 2021-02-02 | 腾讯科技（深圳）有限公司 | Image processing method, image processing apparatus, and computer-readable storage medium |
| CN114972008A (en) * | 2021-11-04 | 2022-08-30 | 华为技术有限公司 | Coordinate restoration method and device and related equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102663361B (en) | 2014-01-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102663361A (en) | Face image reversible geometric normalization method facing overall characteristics analysis | |
| CN107154023B (en) | Based on the face super-resolution reconstruction method for generating confrontation network and sub-pix convolution | |
| CN111047548B (en) | Attitude transformation data processing method and device, computer equipment and storage medium | |
| CN105741252B (en) | Video image grade reconstruction method based on rarefaction representation and dictionary learning | |
| CN101877143B (en) | Three-dimensional scene reconstruction method of two-dimensional image group | |
| CN103345736B (en) | A kind of virtual viewpoint rendering method | |
| KR101028628B1 (en) | Image texture filtering method, storage medium of storing program for executing the same and apparatus performing the same | |
| Kasem et al. | Spatial transformer generative adversarial network for robust image super-resolution | |
| CN112906675B (en) | Method and system for detecting non-supervision human body key points in fixed scene | |
| CN102402784A (en) | Human face image super-resolution method based on nearest feature line manifold learning | |
| CN116958437A (en) | Multi-view reconstruction method and system integrating attention mechanism | |
| Zhang et al. | Color-guided depth image recovery with adaptive data fidelity and transferred graph Laplacian regularization | |
| CN107491747A (en) | Face Forecasting Methodology based on regression analysis and wavelet transformation | |
| Mishra et al. | Self-FuseNet: data free unsupervised remote sensing image super-resolution | |
| Ma et al. | Recovering realistic details for magnification-arbitrary image super-resolution | |
| Almasri et al. | Rgb guided thermal super-resolution enhancement | |
| CN110310269B (en) | Light field image quality evaluation method based on polar plane multi-scale Gabor characteristic similarity | |
| Cheng et al. | Face super-resolution through dual-identity constraint | |
| CN109559278B (en) | Super resolution image reconstruction method and system based on multiple features study | |
| CN107481196B (en) | Feature transformation face super-resolution reconstruction method based on nearest feature line | |
| Zhuang et al. | Dimensional transformation mixer for ultra-high-definition industrial camera dehazing | |
| Li et al. | Point-Based Neural Scene Rendering for Street Views | |
| Wang | Single image super-resolution with u-net generative adversarial networks | |
| CN115272450A (en) | Target positioning method based on panoramic segmentation | |
| CN105096352A (en) | Significance-driven depth image compression method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20211020 Address after: 261021 Room 201, 2 / F, 68 creative space, 200m West Road, intersection of Fushou West Street and Heping Road, Weicheng District, Weifang City, Shandong Province Patentee after: Shandong Wangyuan Information Technology Co.,Ltd. Address before: 100124 No. 100 Chaoyang District Ping Tian Park, Beijing Patentee before: Beijing University of Technology |
|
| TR01 | Transfer of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20140101 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |