CN101105858A - Chinese character composition and realization method based on optimum affine conversion - Google Patents
Chinese character composition and realization method based on optimum affine conversion Download PDFInfo
- Publication number
- CN101105858A CN101105858A CNA2007100290790A CN200710029079A CN101105858A CN 101105858 A CN101105858 A CN 101105858A CN A2007100290790 A CNA2007100290790 A CN A2007100290790A CN 200710029079 A CN200710029079 A CN 200710029079A CN 101105858 A CN101105858 A CN 101105858A
- Authority
- CN
- China
- Prior art keywords
- affine transformation
- chinese character
- image
- optimal
- basic
- 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
Abstract
The invention provides a Chinese compounding realization method based on the optimum global affine conversion. The method adopts a basic Chinese part library to store part images of composing Chinese characters, and convert the parts in the basic Chinese part library through affine conversion parameters obtained through global affine conversion, and then construct converted parts into Chinese characters. The invention designs 55 basic Chinese part images, offers the calculation method of global affine conversion parameters, and judges the structure quality of compounding Chinese images with the method of structure similarity. By the 55 basic Chinese part images and the corresponding affine conversion, the invention can construct Chinese characters at will to greatly lower the storage volume of Chinese characters.
Description
Technical Field
The invention belongs to the field of Chinese character synthesis of information technology, and particularly relates to a Chinese character synthesis implementation method based on optimal global affine transformation.
Technical Field
Chinese characters in the information age expose the weaknesses of large memory capacity, which make the use of Chinese character word stock in various electronic products difficult. The adoption of a hierarchical Chinese character library to reduce the storage capacity of Chinese character images is a feasible solution. In the Chinese character image storage, each Chinese character is composed of several components including basic strokes, radicals such as a vertical horizontal vertical downward and horizontal downward stroke 20058, etc., radicals such as radicals { character radicals } 29357, etc., characters such as Yu Gui Bing Tian exemption, etc., many parts are frequently repeated in different characters, so that only the commonly used parts are stored in a basic Chinese character part library, and each Chinese character is generated by the parts, thereby achieving the aim of reducing the storage amount. However, the same part presents different forms in different Chinese characters, for example, "25909" in "Ao, you," wood "in" forest "," tree "," branch "," fruit ", how to design the hierarchical Chinese character library and how to transform parts in the library makes it important to generate these different forms.
The prior art includes dividing a Chinese character into several blocks, and then splicing the parts according to the proportion between the blocks; the method also comprises the steps of splicing various components according to the structures among the Chinese character components and adjusting the proportional structure of the components so as to achieve better effect. These methods all use the concept of parts, but the use of the splicing method in generating Chinese characters makes the generated Chinese characters rather rigid.
When a Chinese character is generated by using components, the components need to be translated, zoomed and deformed, so that an affine transformation method using manual point selection is also adopted in the prior art to construct a basic Chinese character component library, namely, corresponding three points are manually selected on the Chinese character and the components so as to obtain required affine transformation parameters. However, only the first-level word stock of the national standard of the 'regular script GB _ 2312' contains 3755 Chinese characters and more than five hundred parts in the word stock, and if points are manually selected, the manual workload is huge, the influence of subjective factors is large, the accuracy is low, and time and labor are consumed.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and the optimal global affine transformation method is used for constructing the basic Chinese character component library, so that the geometric shapes and positions of components in the character library and corresponding components in Chinese characters can be automatically matched, affine transformation parameters are calculated, the efficiency is improved on the basis of ensuring a simulation result, and the speed of calculating affine transformation parameters is also improved.
In order to realize the purpose of the invention, the technical scheme is as follows:
the method comprises the steps of designing and storing component images forming the Chinese characters by adopting a basic Chinese character component library, transforming components in the basic Chinese character component library by utilizing affine transformation parameters obtained by global affine transformation, and forming the transformed components into the Chinese characters. The basic Chinese character component library is composed of 55 basic Chinese character component images as follows:
s is used for representing components in the basic Chinese character component library, namely a source image, R is used for representing corresponding components of a standard Chinese character image, namely a target image:
S＝{s 1 ，s 2 ，...，s i ，...，s m }
R＝{r 1 ，r 2 ，...，r j ，...，r n ) (1)
s i is the ith black pixel point of S, r j Is the jth black pixel point of R, and the arrangement order of the points may be arbitrary. Then s i Point s after global affine transformation i * Comprises the following steps:
a and b are affine transformation parameters, A is a two-dimensional matrix, b is a two-dimensional column vector, and S is subjected to overall affine transformation to obtain a synthetic Chinese character image
Make S * Weighted mean of nearest neighbor distances D of sum R NN Minimization is performed to solve for affine transformation parameters A and b, the mathematical representation being shown in equation (2)
→Min for A，b
Wherein the content of the first and second substances,
solving equation (2) only needs phi to differentiate A and b simultaneously to obtain equation set shown in equation (3) and equation (4):
and transposing both sides of the formula (3) simultaneously to obtain a formula (5):
the system of equations shown in equation (6) is composed of equations (4) and (5):
solving equation set (6) results in affine transformation parameters A, b as in equation (7):
wherein, each parameter in the formula (7) is as follows:
the calculation process of the affine transformation parameters also comprises the calculation of optimal affine transformation parameters, the optimal affine transformation parameter calculation adopts an iterative process to obtain optimal A and optimal b, and the iterative process adopts S * Replacing the source image S in the formula (2), when S * Ending when the distance from the target image R in the basic Chinese character component library is not reduced any more. Optimal affine transformation parameters a GAT And b GAT The specific iterative calculation process of (2) is as follows:
suppose that the affine transformation parameter obtained in the ith iteration process is A i And b i The total affine transformation parameter after the ith iteration is A i GAT And b i GAT Then the image S after the nth iteration n * The following were used:
at the same time, the user can select the desired position,
from equations (8) and (9), the total affine transformation parameters after the nth iteration are:
when n =1
In order to obtain a better transformation effect, a part source image and a target image in a basic Chinese character part library are preprocessed before affine transformation parameters are calculated, wherein the preprocessing comprises aligning the gravity centers of the two images, extracting a contour, a skeleton or feature points, and the feature points comprise inflection points and end points on the contour. Affine transformation is only carried out on the feature points, so that the computational complexity is effectively reduced. The use of the center of gravity alignment method may improve simulation quality. The invention uses the structure similarity evaluation method which takes the image structure as the evaluation center to judge whether the structural relationship of each part of the composite Chinese character is reasonable and whether the composite Chinese character is beautiful. The structure similarity evaluation method is specifically realized as follows:
respectively using X = { X = i I =1,2,. Cndot., N } and Y = { Y = i I =1, 2., N } represents the source image and the test image, and the similarity Q is defined as shown in equation (11):
wherein the content of the first and second substances,
q has a value range of [ -1,1]The larger the value is, the higher the structural similarity of the two images is.
The image quality of different areas differs for the same image, so that a sliding window is used,if the number of sliding windows is M, the structural similarity in each window is Q j And then, the final structural similarity of the whole image is as follows:
the invention is based on the transformation relation between the basic Chinese character component library and the real Chinese character constructed by the optimal global affine transformation method, is beneficial to automatically generating the Chinese characters by the limited basic component library and affine transformation parameters, greatly reduces the storage amount of the character library, has wide application value, and can greatly reduce the cost required by the storage amount when being applied to a palm computer, a mobile phone, an embedded system or a singlechip.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention;
fig. 2 is a flowchart of calculating the similarity.
Detailed Description
The invention adopts the basic Chinese character component library to store the component images forming the Chinese characters, and utilizes affine transformation parameters obtained by global imitative transformation to transform the components in the basic Chinese character component library, and then forms the transformed components into the Chinese characters. The basic Chinese character component library is composed of 55 basic Chinese character component images as follows:
s is used for representing components in the basic Chinese character component library, namely a source image, R is used for representing corresponding components of a standard Chinese character image, namely a target image:
S＝{s 1 ，s 2 ，...，s i ，...，s m }
R＝{r 1 ，r 2 ，...，r j ，...，r n } (1)
s i is the ith black pixel point of S, r j The j-th black pixel point of R may be arranged in any order. Then s i Corresponding point S after global affine transformation i * Comprises the following steps:
a and b are affine transformation parameters, A is a two-dimensional matrix, b is a two-dimensional column vector, and S is subjected to overall affine transformation to obtain a synthetic Chinese character image
Make S * Weighted mean of nearest neighbor distances D of sum R NN Minimizing to solve for affine transformation parameters A and b, the mathematical representation is as shown in equation (2)
→Min forA，b
Wherein, the first and the second end of the pipe are connected with each other,
solving equation (2) only needs phi to differentiate A and b simultaneously to obtain equation set shown in equation (3) and equation (4):
and transposing both sides of the formula (3) simultaneously to obtain a formula (5):
the system of equations shown in equation (6) is composed of equations (4) and (5):
solving equation set (6) results in affine transformation parameters A, b as in equation (7):
wherein, each parameter in the formula (7) is as follows:
the above affine transformation parameter calculation process further comprises the optimal affine transformation parameter calculationThe affine transformation parameter calculation adopts an iterative process to obtain optimal A and optimal b, and the iterative process adopts S * Replacing the source image S in the formula (2), when S * Ending when the distance from the target image R in the basic Chinese character component library is not reduced any more. Optimal affine transformation parameter A GAT And b GAT The specific iterative calculation process of (2) is as follows:
suppose that the affine transformation parameter obtained in the ith iteration process is A i And b i The total affine transformation parameter after the ith iteration is A i GAT And b i GAT Then the image S after the nth iteration n * The following:
at the same time, the user can select the desired position,
the total affine transformation parameters after the nth iteration can be obtained from equations (8) and (9) as follows:
when n =1
The method comprises the steps of preprocessing a part source image and a target image in a basic Chinese character part library before calculating affine transformation parameters, wherein the preprocessing comprises extracting contours, skeletons or feature points, and the feature points comprise inflection points and end points on the contours. Because GAT transformation needs to traverse each point on the Chinese character image, in the embodiment, the source image and the target image are preprocessed before affine transformation parameters are calculated: extracting outlines, skeletons or characteristic points (inflection points and end points on the outlines), wherein the basic shapes of Chinese characters and parts are not changed by the preprocessing, so that affine transformation parameters are not greatly influenced, and the speed of calculating the parameters is improved, and the method is called SGAT.
The result of synthesizing Chinese characters by adopting the SGAT is not ideal, and particularly, the SGAT basically has no simulation capability on single strokes, so that some improvements on the SGAT are needed. In the present embodiment, in the SGAT iteration process, a process is added to a source image and a target image: the centers of gravity of the two images are aligned. I.e. preprocessing, includes extracting contours, skeletons or feature points, including inflection points and end points on the contours, aligning the center of gravity of the two images. In this embodiment, a method of aligning the centers of gravity of two images is referred to as IGAT, and after the centers of gravity of the images are aligned by IGAT, the correct matching rate of the chinese character and the nearest neighboring point of the component is improved, so that a better affine transformation parameter can be obtained. Therefore, in order to obtain the optimal affine transformation parameters for each Chinese character, the optimal parameters are selected from the parameters obtained by SGAT (unaligned image barycenter) and IGAT (aligned image barycenter) in each iteration process, the method is called as NGAT herein, as shown in fig. 2, after the pre-processing of extracting contours, skeletons or feature points is carried out on an original image and a target image, affine transformation parameters are respectively calculated through SGAT and IGAT, and then the optimal parameters are selected in the iteration process according to different Chinese characters.
In the embodiment, a structural similarity evaluation method is further used for judging whether the structural relationship of each component of the combined Chinese character is reasonable and whether the combined Chinese character is beautiful, and the structural similarity evaluation method takes the image structure as an evaluation center.
The structure similarity evaluation method is specifically realized as follows:
respectively using X = { X i I =1,2,. N } and Y = { Y = | i = i I =1, 2.. N } represents the source image and the test image, and the definition of the similarity Q is shown in formula (11):
wherein the content of the first and second substances,
q has a value range of [ -1,1]When y is i ＝x i (i =1, 2.. N), Q =1; when in use
Q = -1;
to understand the meaning of Q more clearly, Q is simply rewritten:
the first term of Q represents a linear relationship between X and Y when Y i ＝ax i When + b (i =1, 2.. N, a, b are constants and a > 0), the term is 1; the second term represents the luminance relationship between X and Y when
When this term equals 1; the third term represents the similarity between X and Y when σ is x ＝σ y When this term equals 1;
the image quality of different areas is different for the same image, so that sliding windows are used, and if the number of sliding windows is M, the structural similarity in each window is Q j Then, the final structural similarity of the whole image is:
the definition formula (12) of Q contains a denominator, and when a sliding window is used for the binary Chinese character, the denominator of Q is likely to be 0, so the invention is derived aiming at the calculation problem of Q:
(1) As long as the feature of the image is not represented by "0", the feature of the image in the formula (11) can be ensured
(2) When the temperature is higher than the set temperature
Time, σ x ＝σ y =0, the third term in equation (12) is equal to 1. According to σ x 2 And σ y 2 Can be defined at this time
If it is not
X is then i ＝y i So Q =1; if it is not
Let a =1 and a be set in a,
then y is i ＝ax i + b, so the first term of equation (12) is 1, so Q is now equal to the second term of equation (12), and the flow chart is shown in fig. 2.
This embodiment compares various affine transformation methods and the time required for the pre-processing to calculate the affine transformation parameters, as shown in table 1:
table 1 (in ms): time taken for various affine transformations and preprocessing calculation parameters (SPH: manual Point selection, C: contour, S: skeleton, F: feature points)
| SPH | SGAT (C) | SGAT (S) | SGAT (F) | NGAT (C) | NGAT (S) | NGAT (F) | |
| Striped mullet | 6110 | 22155 | 6812 | 218 | 12156 | 2765 | 187 |
| Reeling machine | 10203 | 34000 | 4609 | 156 | 21454 | 5687 | 204 |
| Invasion of | 968 | 13889 | 3828 | 189 | 6562 | 1671 | 218 |
| Plant | 1670 | 20939 | 6140 | 203 | 7391 | 2219 | 203 |
| Wiping cloth | 2173 | 6595 | 2188 | 188 | 3407 | 845 | 266 |
| Average out | 4224.8 | 19515.6 | 4715.4 | 190.8 | 10194 | 2637.4 | 215.6 |
As can be seen from table 1:
(1) For three different pre-treatments (contours, skeletons and feature points), feature points have distinct advantages, whether SGAT or NGAT. The characteristic points are far less than the points of the outline and the skeleton, so that the calculation complexity is greatly reduced, and the calculation speed is improved.
(2) When using the feature points, the time used by the SGAT and the NGAT is much less than the time of manual selection, even by as much as ten times.
(3) When feature points are employed, NGAT has comparable temporal performance to SGAT. But the simulated effect of NGAT is incomparable by SGAT.
In this embodiment, a 4 × 4 sliding window is used when calculating the structural similarity between the synthesized chinese characters and the standard chinese characters, and 3755 synthesized chinese characters of the "regular script GB _2312" in the national standard first-level word library of the manual point selection method are evaluated first, and the chinese characters are ranked from good to bad according to the simulation effect, in order to explain the effectiveness of the NGAT, 38 chinese characters are taken altogether from 1900 characters with a better simulation effect in the front by taking 50 chinese characters as intervals, 74 chinese characters are taken altogether from 25 characters with a poorer simulation effect in the back by taking intervals, and finally, 117 chinese characters are obtained as experimental samples by the 5 worst total selections, and the average structural similarity of the various methods of these samples is shown in table 2:
table 2: average structural similarity of various methods
| Manual point selection | NGAT Characteristic point | NGAT Contour profile | NGAT Skeleton |
| 0.8945 | 0.8926 | 0.8727 | 0.8830 |
It can be seen from table 2 that when feature points are used, the average structural similarity of NGAT is substantially equal to that of manual point selection, but automatic point matching and high calculation speed of NGAT are not achievable by the manual point selection method.
The invention utilizes the affine transformation parameter obtained by GAT to transform the components in the hierarchical word stock, and the components form Chinese characters, and uses the image structure as the structural similarity evaluation method of the evaluation center, rather than the simple comparison of pixel points, which is more suitable for evaluating the composite Chinese characters formed by the hierarchical word stock. The invention can synthesize any Chinese character by utilizing 55 basic component images and corresponding affine transformation, only the 55 basic component images and affine transformation parameters need to be stored in practical application, and the images of each Chinese character do not need to be stored, so that the Chinese character storage capacity is greatly reduced.
Claims (10)
1. A Chinese character synthesizing method based on optimal global affine transformation is characterized in that a basic Chinese character component library is used for storing component images forming Chinese characters, affine transformation parameters obtained through global affine transformation are used for transforming components in the basic Chinese character component library, and then the transformed components form the Chinese characters.
2. The method for implementing optimal global affine transformation based on Chinese character synthesis according to claim 1, wherein said global affine transformation process is as follows:
expressing the components in the basic Chinese character component library by S, namely the source image, and expressing the corresponding components of the standard Chinese character image by R, namely the target image:
S＝{s 1 ，s 2 ，...，s i ，...，s m }
R＝{r 1 ，r 2 ，...，r j ，...，r n }
let s i Is the ith black pixel point of S, r j Is the jth black pixel point of R, then s i Corresponding point s after global refraction mapping i * Comprises the following steps:
a and b are affine transformation parameters, A is a two-dimensional matrix, b is a two-dimensional column vector, and the obtained composite Chinese character image after S is subjected to global affine transformation is
3. The method for implementing optimal global affine transformation based on Chinese character synthesis according to claim 1, wherein said basic Chinese character component library comprises the following 55 basic Chinese character components:
4. the method for implementing Chinese character synthesis based on optimal global affine transformation as claimed in claim 2, wherein the parameter calculation process of the global affine transformation is as follows:
make S * Weighted mean of nearest neighbor distances D of sum R NN Minimization is performed to solve for affine transformation parameters A and b, the mathematical representation being shown in equation (2)
Wherein the content of the first and second substances,
solving equation (2) only needs phi to differentiate A and b simultaneously to obtain equation set shown in equation (3) and equation (4):
and transposing both sides of the formula (3) simultaneously to obtain a formula (5):
the system of equations shown in equation (6) is composed of equations (4) and (5):
solving equation set (6) results in affine transformation parameters A, b as in equation (7):
wherein, each parameter in the formula (7) is as follows:
5. the method for implementing Chinese character synthesis based on optimal global affine transformation as claimed in claim 2, wherein the affine transformation parameter calculation process further includes optimal affine transformation parameter calculation, the optimal affine transformation parameter calculation adopts an iterative process to obtain optimal A and optimal b, and the iterative process adopts S * Replacing the source image S in equation (2), when S * Ending when the distance from the target image R in the basic Chinese character component library is not reduced any more.
6. The method of claim 5, wherein said optimal affine transformation parameters A GAT And b GAT The specific iterative calculation process of (2) is as follows:
the affine transformation parameter obtained in the ith iteration process is assumed to be A i And b i The total affine transformation parameter after the ith iteration is A i GAT And b i GAT Then the image S after the nth iteration n * The following:
at the same time, the user can select the desired position,
the total affine transformation parameters after the nth iteration can be obtained from equations (8) and (9) as follows:
when n =1
7. The method for realizing Chinese character synthesis based on optimal global affine transformation according to claim 6, wherein before affine transformation parameters are calculated, preprocessing is performed on part source images and target images in a basic Chinese character part library, the preprocessing comprises extracting contours, skeletons or feature points, and the feature points comprise inflection points and end points on the contours.
8. The method for implementing Chinese character synthesis based on optimal global affine transformation according to claim 6, characterized in that the parts source image and target image in the basic Chinese character parts library are preprocessed in the iterative process of calculating affine transformation parameters, wherein the preprocessing refers to aligning the barycenter of two images.
9. The method for implementing optimal global affine transformation-based Chinese character synthesis according to claim 7 or 8, further comprising using a structural similarity evaluation method with an image structure as an evaluation center to evaluate whether structural relationships of components constituting the synthesized Chinese character are reasonable and whether the synthesized Chinese character is beautiful.
10. The method for implementing Chinese character synthesis based on optimal global affine transformation according to claim 9, wherein said method for evaluating structural similarity is implemented as follows:
respectively using X = { X = i I =1,2,. Cndot., N } and Y = { Y = i I =1,2,..,. N } represents the source chinese character image and the target image, and the similarity Q is defined as shown in formula (11):
wherein, the first and the second end of the pipe are connected with each other,
q has a value in the range of [ -1,1]；
And the sliding windows are used for representing the difference of image quality of different areas of the same image, if the number of the sliding windows is M, the structural similarity in each window is Q j And then the final structural similarity of the whole image is as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2007100290790A CN100533478C (en) | 2007-07-09 | 2007-07-09 | Chinese character composition and realization method based on optimum affine conversion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2007100290790A CN100533478C (en) | 2007-07-09 | 2007-07-09 | Chinese character composition and realization method based on optimum affine conversion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101105858A true CN101105858A (en) | 2008-01-16 |
| CN100533478C CN100533478C (en) | 2009-08-26 |
Family
ID=38999741
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2007100290790A Expired - Fee Related CN100533478C (en) | 2007-07-09 | 2007-07-09 | Chinese character composition and realization method based on optimum affine conversion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100533478C (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107241100A (en) * | 2016-03-29 | 2017-10-10 | 北大方正集团有限公司 | Character library component compresses method and device |
| CN107818544A (en) * | 2016-09-12 | 2018-03-20 | 北京大学 | A kind of character scale method |
| CN108305209A (en) * | 2017-01-12 | 2018-07-20 | 富士通株式会社 | Character deformation method and Character deformation equipment |
| CN112434763A (en) * | 2020-11-24 | 2021-03-02 | 伍曙光 | Chinese character skeleton generating method based on computer |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100219072B1 (en) * | 1996-04-02 | 1999-09-01 | 김영환 | Font transformation and rasterizing method using medial axis transform |
| JP3563891B2 (en) * | 1996-10-24 | 2004-09-08 | キヤノン株式会社 | Character generation method and device |
| GB2354099B (en) * | 1999-09-09 | 2003-09-10 | Sony Uk Ltd | Image identification apparatus and method of identifying images |
| SE521911C2 (en) * | 2001-01-15 | 2003-12-16 | Decuma Ab Ideon Res Park | Method, device and computer program for recognizing a handwritten character |
-
2007
- 2007-07-09 CN CNB2007100290790A patent/CN100533478C/en not_active Expired - Fee Related
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107241100A (en) * | 2016-03-29 | 2017-10-10 | 北大方正集团有限公司 | Character library component compresses method and device |
| CN107818544A (en) * | 2016-09-12 | 2018-03-20 | 北京大学 | A kind of character scale method |
| CN107818544B (en) * | 2016-09-12 | 2020-05-05 | 北京大学 | Character scaling method |
| CN108305209A (en) * | 2017-01-12 | 2018-07-20 | 富士通株式会社 | Character deformation method and Character deformation equipment |
| CN108305209B (en) * | 2017-01-12 | 2021-06-04 | 富士通株式会社 | Character deformation method and character deformation apparatus |
| CN112434763A (en) * | 2020-11-24 | 2021-03-02 | 伍曙光 | Chinese character skeleton generating method based on computer |
Also Published As
| Publication number | Publication date |
|---|---|
| CN100533478C (en) | 2009-08-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xia et al. | Efficient non-local contrastive attention for image super-resolution | |
| CN108304357B (en) | Chinese character library automatic generation method based on font manifold | |
| CN103049892B (en) | Non-local image denoising method based on similar block matrix rank minimization | |
| CN105844635B (en) | A kind of rarefaction representation depth image method for reconstructing based on structure dictionary | |
| CN109446185B (en) | Collaborative filtering missing data processing method based on user clustering | |
| CN111507357B (en) | Defect detection semantic segmentation model modeling method, device, medium and equipment | |
| US10169908B2 (en) | Method, apparatus, storage medium and device for controlled synthesis of inhomogeneous textures | |
| CN104077742B (en) | Human face sketch synthetic method and system based on Gabor characteristic | |
| CN106886793B (en) | Hyperspectral image waveband selection method based on discrimination information and manifold information | |
| CN101105858A (en) | Chinese character composition and realization method based on optimum affine conversion | |
| CN102013106A (en) | Image sparse representation method based on Curvelet redundant dictionary | |
| CN115147456A (en) | Target tracking method based on time sequence adaptive convolution and attention mechanism | |
| CN116563355A (en) | Target tracking method based on space-time interaction attention mechanism | |
| CN116168011A (en) | Multi-granularity network-based reference-free image quality evaluation method | |
| CN115965789A (en) | Scene perception attention-based remote sensing image semantic segmentation method | |
| Muralikrishnan et al. | GLASS: Geometric latent augmentation for shape spaces | |
| CN113822825A (en) | Optical building target three-dimensional reconstruction method based on 3D-R2N2 | |
| Guo et al. | Accelerating patch-based low-rank image restoration using kd-forest and Lanczos approximation | |
| CN104809478A (en) | Image partitioning method and device oriented to large-scale three-dimensional reconstruction | |
| Zou et al. | A nonlocal low-rank regularization method for fractal image coding | |
| CN113658130A (en) | No-reference screen content image quality evaluation method based on dual twin network | |
| CN117011655A (en) | Adaptive region selection feature fusion based method, target tracking method and system | |
| CN108876711B (en) | Sketch generation method, server and system based on image feature points | |
| CN106651770A (en) | Method for reconstructing multispectral super-resolution imaging based on Lapras norm regularization | |
| CN109035318B (en) | Image style conversion 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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20090826 Termination date: 20160709 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |