CN108537234A - The demarcation method of rule background dictionary-based learning and rectangular target image - Google Patents
The demarcation method of rule background dictionary-based learning and rectangular target image Download PDFInfo
- Publication number
- CN108537234A CN108537234A CN201810238135.XA CN201810238135A CN108537234A CN 108537234 A CN108537234 A CN 108537234A CN 201810238135 A CN201810238135 A CN 201810238135A CN 108537234 A CN108537234 A CN 108537234A
- Authority
- CN
- China
- Prior art keywords
- matrix
- row
- target image
- dictionary
- rule
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
- G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Abstract
The present invention discloses a kind of demarcation method of rule background dictionary-based learning and rectangular target image：(1) RGB pictures are converted to gray scale picture to store；(2) it is sampled respectively using the priori fringe region horizontal and vertical to picture；(3) lateral, longitudinal sample is trained respectively using dictionary learning algorithm；(4) geometric position of horizontal boundary, longitudinal boundary is obtained using dictionary learning grader；(5) demarcation of rule background and rectangular target image is completed by horizontal, longitudinal boundary geometric position.The classification and Detection of the present invention combination dictionary learning model and coding specification device, classified to dictionary matrix and sparse matrix using least squares estimate, and delimited for rule background in general picture and the separation of rectangular target image, the intelligence of picture is delimited and image rule background separation has great significance and influences.
Description
Technical field
The new demarcation method based on dictionary learning and coding specification device that the present invention relates to a kind of, more particularly to one kind are based on
The rule background of dictionary learning and the demarcation method of rectangular target image.
Background technology
The camera function of becoming increasingly popular and being widely used with mobile terminal, mobile terminal is also generalized out constantly
Come.Integrated with mobile terminal device and imaging sensor, this makes the information interchange in the world become increasingly convenient and efficient.
While the sensor technology of mobile terminal constantly develops and improves, adopting in real time for picture is carried out in information of mobile terminal exchange
Collection and transmission have obtained more and more approvals, and by the use of extensive input reality.
Currently, in pictorial information exchange, there is much the case where using rule striped as background, the presence of these backgrounds makes
The acquisition needs for obtaining target information are delimited by artificial mode, and law of segregation background and target image are usually using SIFT
Algorithm i.e. its derivative algorithm or certain machine learning algorithms for signal specific.
It takes pictures for mobile terminal device, using rule striped as the rectangular target image of background, wherein rectangular target image
Rotation angle at 10 ° and within.For using rule striped as the acquisition of the target image information of background, manually determining if used
The mode on boundary needs the input for greatly increasing cost of labor；And use SIFT and its derivative algorithm or some other machines
The algorithm of study then needs to train the long time, or cannot reach using required precision.
Invention content
In view of the deficiencies of the prior art, a kind of rule background dictionary-based learning of present invention offer and rectangular target image
Demarcation method, in conjunction with the classification and Detection of dictionary learning model and coding specification device, using least squares estimate to dictionary square
Battle array and sparse matrix are classified, and are delimited for the separation of rule background and target image in general picture, for picture
Intelligence is delimited and image background separation has great significance and influences.
The present invention uses following technical scheme to solve above-mentioned technical problem：
The present invention provides a kind of demarcation method of rule background dictionary-based learning and rectangular target image, specifically includes
Following steps：
Step 1, it converts RGB image to gray level image and stores；
Step 2, using priori, horizontal and vertical sampling is carried out to the edge of gray level image respectively, it is horizontal as rule background
To sample and longitudinal sample；Horizontal and vertical sampling is carried out to the center of gray level image, as target image transverse direction sample and is indulged
To sample；
Step 3, using dictionary learning algorithm, respectively to rule background and target image transverse direction sample, rule background and mesh
Logo image longitudinal direction sample is trained；
Step 4, using dictionary learning grader, obtain rule background, target image horizontal and vertical boundary；
Step 5, by rule background, the geometric position on the horizontal and vertical boundary of target image, complete rule background and
The demarcation of target image.
As the further technical solution of the present invention, the storage form of gray level image is elemental range 0-255's in step 1
Matrix.
As the further technical solution of the present invention, gray level image matrix is carried out in the form of vectors respectively in step 2 whole
Row and permutation sampling：The lateral sample of rule background takes the starting N rows of gray level image matrix and end N rows, longitudinal sample to take ash
Spend the starting N row and end N row of image array；The lateral sample of target image takes the centre 2N rows of gray level image matrix, indulges
The centre 2N row of gray level image matrix are taken to sample；Wherein, N is setting value.
As the further technical solution of the present invention, the object function of dictionary learning model is in step 3：
In formula, C=1,2, C=1 indicate to belong to rule background classes, and C=2 indicates rectangular target image class, k=1,2,Table
Show the transposed matrix for the longitudinal sample matrix for belonging to rule background classes,Indicate the lateral sample matrix for belonging to rule background classes
Transposed matrix,Indicate the transposed matrix for belonging to longitudinal sample matrix of rectangular target image class,Expression belongs to rectangle mesh
The transposed matrix of the lateral sample matrix of logo image class,It indicatesCorresponding rarefaction representation matrix, D1Indicate longitudinal sample pair
The dictionary matrix answered, D2Indicate that the corresponding dictionary matrix of lateral sample, β indicate regularization coefficient.
As the present invention further technical solution,Total 2K rows, preceding K rows correspond to the 1st class, and K+1 rows to 2K rows correspond to
2nd class；DkTotal 2K row, preceding corresponding 1st class of K row, K+1 are arranged to 2K and are arranged corresponding 2nd class.
As the further technical solution of the present invention, dictionary matrix DkTraining process it is as follows：
First, matrix D is initialized using random number methodkWith
Secondly, by least square method, matrix D is acquiredkWithIterative formula be：
Wherein,Indicate XkLeast-squares estimation value, I indicate identical transformation matrix, diIndicate DkI-th row, diMost
Small two multiply estimated value It indicatesThe i-th row, LCIndicate DkIn all row or X for belonging to C classeskIn belong to C classes row set；Indicate institute
There is the D for being not belonging to C classeskEach row and XkIn corresponding row the sum of products,It indicates all belong to C classes, remove
DkThe i-th outer D of rowkEach row and XkIn corresponding row the sum of products；
Finally, it using above-mentioned iterative formula, is iterated according to setting iterations, obtains the dictionary matrix of training completion
Dk。
As the further technical solution of the present invention, step 4 is specially：Using dictionary learning grader, to gray level image
The row, column vector of matrix is traversed respectively, obtain rule background, target image horizontal and vertical boundary.
As the further technical solution of the present invention, gray level image matrix column vector is traversed, by below
Formula judges that the dependent of dead military hero is as follows in rule background classes or rectangular target image class, formula：
The transposition of the row vector of gray level image matrix is traversed, judges that the row belongs to the rule back of the body by formula below
Scape class or rectangular target image class, formula are as follows：
Wherein, | |jIndicate the absolute value of the element of column vector jth row, BiIndicate the i-th row of gray level image matrix B,
(BT)iIndicate the i-th row of the transposition of gray level image matrix B；
According to above-mentioned traversal as a result, by geometry sequence obtain the row generic vector of gray level image matrix B, row generic to
Amount becomes 2 by 1 in row generic vector, row generic vector and becomes the position of 1 element in corresponding vector, as target image by 2
Coordinate position where row, column boundary.
The present invention has the following technical effects using above technical scheme is compared with the prior art：Proposed by the invention
The demarcation method of rule background dictionary-based learning and rectangular target image, can be effectively to the picture of mobile terminal shooting
In rule background and rectangular target image carry out demarcation operation, substantially increase the efficiency and accuracy rate of demarcation, utilize priori
Supervision is supervised instead of artificial interaction, also avoids the input of labor cost.
Description of the drawings
Fig. 1 is flow chart of the method for the present invention；
Fig. 2 is the photo of mobile terminal taking under No. 1 rule background；
Fig. 3 is the photo of mobile terminal taking under No. 2 rule backgrounds；
Fig. 4 is the demarcation design sketch of rule background and rectangular target image dictionary-based learning under No. 1 rule background；
Fig. 5 is the demarcation design sketch of rule background and rectangular target image dictionary-based learning under No. 2 rule backgrounds.
Specific implementation mode
Below in conjunction with the accompanying drawings and specific embodiment is described in further detail technical scheme of the present invention：
The problem of present invention is for rule background in differentiation picture and rectangular target image, the present invention proposes priori sampling
With the method for dictionary learning, include the following steps：
(1) fringe region of acquiescence picture is rule background, and the central area of picture is rectangular target image, and to picture
Fringe region and central area be divided into and horizontal and vertical being sampled respectively；
(2) row, column sample training is carried out respectively using the dictionary learning model optimized based on least square method to be corresponded to
Dictionary rectangular array dictionary D1, row dictionary D2。
According to trained dictionary D1、D2, the storage matrix that computer is corresponded to gray scale picture carries out row, column traversal respectively,
By the coding specification device of dictionary learning, determine that the class per a line belongs to, the class ownership of each row, so that it is determined that per a line, it is every
One dependent of dead military hero is in rule background classes or rectangular target image class.
According to above-mentioned traversal as a result, obtain row generic vector, the row generic vector of gray level image matrix by geometry sequence,
In two vectors returned, element can be 1 or 2, wherein the corresponding row or column of 1 expression picture belongs to rule background classes, 2 tables
The corresponding row or column of diagram piece belongs to rectangular target image class.Become 2 by 1 in two return vectors and becomes 1 element by 2,
Position in vector, the as coordinate position where target image row, column boundary, due to gray level image storage mode in computer
It is matrix form, the geometric position on above-mentioned coordinate position namely target image row, column boundary, thus delimit and can complete.
The extraction of dictionary learning sample
For the initial picture of the acquisition of mobile device, we are exercised supervision using priori, and picture is thought in priori supervision
Edge belongs to rule background classes, and the central area of picture belongs to rectangular target image class.After image gray processing, storage form
For the matrix of elemental range 0-255.Full line sampling is carried out to rule background classes and rectangular target image class respectively and permutation is adopted
Sample, for the form of vector.By taking row samples as an example, rule background classes sample takes starting N row vectors, final N row vectors, rectangular target
Image class sample takes centre 2N row vectors, similarly, column vector rule background classes sample take starting N column vectors, final N arrange to
Amount, rectangular target image class sample take centre 2N column vectors, pass through following target formula：
In formula, in formula, C=1,2, C=1 indicate to belong to rule background classes, and C=2 indicates rectangular target image class, k=1,
2,Indicate the transposed matrix for belonging to longitudinal sample matrix of rule background classes,Indicate the lateral sample for belonging to rule background classes
The transposed matrix of this matrix,Indicate the transposed matrix for belonging to longitudinal sample matrix of rectangular target image class,Expression belongs to
The transposed matrix of the lateral sample matrix of rectangular target image class,It indicatesCorresponding rarefaction representation matrix, D1Indicate longitudinal
The corresponding dictionary matrix of sample, D2Indicate that the corresponding dictionary matrix of lateral sample, β indicate regularization coefficient.Total 2K rows, preceding K
Corresponding 1st class of row, K+1 rows to 2K rows correspond to the 2nd class；DkTotal 2K row, preceding corresponding 1st class of K row, K+1 are arranged to 2K to arrange and be corresponded to
2nd class.
The training of dictionary
D is initialized using random number methodkMatrix andMatrix is acquired by least square methodMatrix, DkMatrix changes
It is as follows for formula, wherein DkMatrix is by Leie time update：
Wherein,Indicate XkLeast-squares estimation value,Representing matrix DkTransposition, I indicate identical transformation matrix, di
Indicate DkI-th row,It is diLeast-squares estimation value,It indicatesThe i-th row transposition；Indicate corresponding diResidual error
Matrix,LCIndicate DkIn all row or X for belonging to C classeskIn belong to C classes row set；Table
Show all D for being not belonging to C classeskEach row and XkIn corresponding row the sum of products,It indicates all and belongs to C
Class removes DkThe i-th outer D of rowkEach row and XkIn corresponding row the sum of products.
15-20 iteration is carried out using above formula, obtains the D that finally training is completedkMatrix.
Grader is delimited
According to the row, column sample of gray level image matrix B, wherein row sample transposition is suitable for learning model, instructed by dictionary
White silk can obtain corresponding two dictionaries.After obtaining corresponding dictionary, for the transposition of the row vector of matrix B, column vector into
Row traversal, by taking column vector as an example, judges the dependent of dead military hero of B in rule background classification or rectangular target image class by formula below
Not, formula is as follows：
Wherein, | |jIndicate the absolute value of the element of column vector jth row.BiIndicate the i-th row of B matrixes,It is a column vector, and line number and XkLine number it is consistent, therefore analogy Xk, preceding K rows belong to the first kind, K
+ 1 belongs to the second class to 2K rows,It indicates respectively to the absolute value of the element of of a sort all rows
Summation.
Similarly, its transposition is about to for row vector, the i-th of corresponding B, uses following classification formula：
Using traversal as a result, returning to every a line of B, each dependent of dead military hero respectively in rule background or rectangular target image.
According to above-mentioned traversal as a result, obtain row generic vector, the row generic vector of gray level image matrix by geometry sequence, the two of return
In a vector, element can be 1 or 2, wherein the corresponding row or column of 1 expression picture belongs to rule background classes, 2 indicate picture pair
The row or column answered belongs to rectangular target image class.Become 2 by 1 in two return vectors and become 1 element by 2, in vector
Position, the as coordinate position where target image row, column boundary, since gray level image storage mode is rectangular in computer
The geometric position on formula, above-mentioned coordinate position namely target image row, column boundary, thus delimit and can complete.
Embodiment
We have chosen two kinds of striped backgrounds, and one is the rule backgrounds parallel with target image, and one is inclined stripe rule
Background is restrained, meanwhile, it is suitable for rectangular target image between our inventive method, we choose invoice as target image.It is real
Proved recipe method be rectangle invoice different in size is placed in different rule backgrounds, is taken pictures using mobile phone terminal, and on
The illustration that computer is passed to as us (moves terminal taking under No. 1 rule background and No. 2 rule backgrounds as shown in Figures 2 and 3
Photo), using our inventive method, carry out the demarcation processing of picture, the result of demarcation is as shown in Figure 4 and Figure 5.According to fixed
The display result on boundary, it was demonstrated that our inventive method fully achieves very high accuracy and practical standard.
The present invention realizes on MATLAB experiment porch, as shown in Figure 1, including mainly several steps：
1st step：A certain rectangular target image is shot in rule background using mobile terminal device.Execute the 2nd
Step；
2nd step：Picture is imported into MATLAB, and is processed into double data type matrixes.Execute the 3rd step；
3rd step：It being sampled using priori supervision, edge is rule background sample, and center is rectangular target image pattern,
Execute the 4th step；
4th step：The training of sample is carried out according to formula (2) (3) (4), iteration 15-20 times or so executes the 5th step；
5th step：The traversal for being carried out two dimensions of portraitlandscape to picture matrix using formula (5), is determined belonging to vector
Class returns to the vector of row, column generic, executes the 6th step.
6th step：Using difference reject return vector wrong data, obtain rule background and rectangular target image row,
Arrange the geometric position information on boundary.
The present invention discloses a kind of demarcation method of rule background dictionary-based learning and rectangular target image.This method walks
Suddenly：(1) RGB pictures are converted to gray scale picture to store；(2) fringe region point for utilizing priori horizontal and vertical to picture
It does not carry out sampling sampling respectively lateral as the central area of rule background sample and picture, longitudinal and is used as rectangular target image
Sample；(3) lateral, the longitudinal sample of rule background sample and rectangular target image is carried out respectively using dictionary learning algorithm
Training；(4) horizontal boundary of rule background and rectangular target image, the geometry of longitudinal boundary are obtained using dictionary learning grader
Position；(5) demarcation of rule background and rectangular target image is completed by horizontal, longitudinal boundary geometric position.The present invention combines
The classification and Detection of dictionary learning model and coding specification device carries out dictionary matrix and sparse matrix using least squares estimate
Classification, and being delimited for rule background in general picture and the separation of rectangular target image, the intelligence of picture is delimited and
Image rule background separation has great significance and influences.
The above, the only specific implementation mode in the present invention, but scope of protection of the present invention is not limited thereto, appoints
What is familiar with the people of the technology within the technical scope disclosed by the invention, it will be appreciated that expects transforms or replaces, and should all cover
Within the scope of the present invention, therefore, the scope of protection of the invention shall be subject to the scope of protection specified in the patent claim.
Claims (8)
1. the demarcation method of rule background dictionary-based learning and rectangular target image, which is characterized in that specifically include following
Step：
Step 1, it converts RGB image to gray level image and stores；
Step 2, using priori, horizontal and vertical sampling is carried out to the edge of gray level image respectively, as rule background transverse direction sample
Sheet and longitudinal sample；Horizontal and vertical sampling is carried out to the center of gray level image, as target image transverse direction sample and longitudinal sample
This；
Step 3, using dictionary learning algorithm, respectively to rule background and target image transverse direction sample, rule background and target figure
As longitudinal sample is trained；
Step 4, using dictionary learning grader, obtain rule background, target image horizontal and vertical boundary；
Step 5, by rule background, the geometric position on the horizontal and vertical boundary of target image, rule background and target are completed
The demarcation of image.
2. the demarcation method of rule background dictionary-based learning and rectangular target image according to claim 1, special
Sign is that the storage form of gray level image is the matrix of elemental range 0-255 in step 1.
3. the demarcation method of rule background dictionary-based learning and rectangular target image according to claim 2, special
Sign is, carries out full line and permutation sampling in step 2 in the form of vectors to gray level image matrix respectively：The lateral sample of rule background
Originally the starting N rows and end N rows of gray level image matrix, longitudinal sample is taken to take the starting N row and end N row of gray level image matrix；
The lateral sample of target image takes the centre 2N rows of gray level image matrix, longitudinal sample to take the centre 2N of gray level image matrix
Row；Wherein, N is setting value.
4. the demarcation method of rule background dictionary-based learning and rectangular target image according to claim 3, special
Sign is that the object function of dictionary learning model is in step 3：
In formula, C=1,2, C=1 indicate to belong to rule background classes, and C=2 indicates rectangular target image class, k=1,2,It indicates to belong to
In the transposed matrix of longitudinal sample matrix of rule background classes,Indicate turning for the lateral sample matrix for belonging to rule background classes
Matrix is set,Indicate the transposed matrix for belonging to longitudinal sample matrix of rectangular target image class,Expression belongs to rectangular target figure
As the transposed matrix of the lateral sample matrix of class,It indicatesCorresponding rarefaction representation matrix, D1Indicate that longitudinal sample is corresponding
Dictionary matrix, D2Indicate that the corresponding dictionary matrix of lateral sample, β indicate regularization coefficient.
5. the demarcation method of rule background dictionary-based learning and rectangular target image according to claim 4, special
Sign is,Total 2K rows, preceding K rows correspond to the 1st class, and K+1 rows to 2K rows correspond to the 2nd class；DkTotal 2K row, preceding K row the corresponding 1st
Class, K+1 are arranged to 2K and are arranged corresponding 2nd class.
6. the demarcation method of rule background dictionary-based learning and rectangular target image according to claim 4, special
Sign is, dictionary matrix DkTraining process it is as follows：
First, matrix D is initialized using random number methodkWith
Secondly, by least square method, matrix D is acquiredkWithIterative formula be：
Wherein,Indicate XkLeast-squares estimation value, I indicate identical transformation matrix, diIndicate DkI-th row, diMinimum two
Multiply estimated value It indicatesThe i-th row, LC
Indicate DkIn all row or X for belonging to C classeskIn belong to C classes row set；It indicates all and is not belonging to C classes
DkEach row and XkIn corresponding row the sum of products,It indicates all belong to C classes, remove DkThe i-th outer D of rowk
Each row and XkIn corresponding row the sum of products；
Finally, it using above-mentioned iterative formula, is iterated according to setting iterations, obtains the dictionary matrix D of training completionk。
7. the demarcation method of rule background dictionary-based learning and rectangular target image according to claim 6, special
Sign is that step 4 is specially：Using dictionary learning grader, the transposition of the row vector of gray level image matrix, column vector are distinguished
Traversed, obtain rule background, target image horizontal and vertical boundary.
8. the demarcation method of rule background dictionary-based learning and rectangular target image according to claim 7, special
Sign is, is traversed to gray level image matrix column vector, by formula below judge the dependent of dead military hero in rule background classes or
Person's rectangular target image class, formula are as follows：
The transposition of the row vector of gray level image matrix is traversed, judges that the row belongs to rule background classes by formula below
Or rectangular target image class, formula are as follows：
Wherein, | |jIndicate the absolute value of the element of column vector jth row, BiIndicate the i-th row of gray level image matrix B, (BT)iTable
Show the i-th row of the transposition of gray level image matrix B；
According to above-mentioned traversal as a result, obtaining row generic vector, the row generic vector of gray level image matrix B, row by geometry sequence
Generic vector becomes 2 by 1 in row generic vector and becomes the position of 1 element in corresponding vector, as target image row, column by 2
Coordinate position where boundary.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810238135.XA CN108537234A (en) | 2018-03-22 | 2018-03-22 | The demarcation method of rule background dictionary-based learning and rectangular target image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810238135.XA CN108537234A (en) | 2018-03-22 | 2018-03-22 | The demarcation method of rule background dictionary-based learning and rectangular target image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108537234A true CN108537234A (en) | 2018-09-14 |
Family
ID=63484096
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810238135.XA Pending CN108537234A (en) | 2018-03-22 | 2018-03-22 | The demarcation method of rule background dictionary-based learning and rectangular target image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108537234A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101283364A (en) * | 2005-09-16 | 2008-10-08 | 索尼电子有限公司 | Extracting a moving object boundary |
| CN101582159A (en) * | 2009-06-26 | 2009-11-18 | 哈尔滨工业大学 | Infrared image background suppression method based on unsupervised kernel regression analysis |
| CN107832786A (en) * | 2017-10-31 | 2018-03-23 | 济南大学 | A kind of recognition of face sorting technique based on dictionary learning |
-
2018
- 2018-03-22 CN CN201810238135.XA patent/CN108537234A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101283364A (en) * | 2005-09-16 | 2008-10-08 | 索尼电子有限公司 | Extracting a moving object boundary |
| CN101582159A (en) * | 2009-06-26 | 2009-11-18 | 哈尔滨工业大学 | Infrared image background suppression method based on unsupervised kernel regression analysis |
| CN107832786A (en) * | 2017-10-31 | 2018-03-23 | 济南大学 | A kind of recognition of face sorting technique based on dictionary learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Dong et al. | Remote sensing image super-resolution using second-order multi-scale networks | |
| CN110246181B (en) | Anchor point-based attitude estimation model training method, attitude estimation method and system | |
| CN108764024A (en) | Generating means, method and the computer readable storage medium of human face recognition model | |
| CN108596108B (en) | Aerial remote sensing image change detection method based on triple semantic relation learning | |
| CN108647639A (en) | Real-time body's skeletal joint point detecting method | |
| CN105844630B (en) | A kind of image super-resolution fusion denoising method of binocular vision | |
| CN107680116B (en) | Method for monitoring moving target in video image | |
| CN106023187B (en) | A kind of method for registering images based on SIFT feature and angle relative distance | |
| CN104537705B (en) | Mobile platform three dimensional biological molecular display system and method based on augmented reality | |
| CN105335725A (en) | Gait identification identity authentication method based on feature fusion | |
| Chen et al. | Convolutional neural network based dem super resolution | |
| CN103578093A (en) | Image registration method and device and augmented reality system | |
| CN107301643B (en) | Well-marked target detection method based on robust rarefaction representation Yu Laplce's regular terms | |
| CN111709980A (en) | Multi-scale image registration method and device based on deep learning | |
| CN114332639B (en) | Satellite attitude vision measurement method of nonlinear residual error self-attention mechanism | |
| CN113111758B (en) | SAR image ship target recognition method based on impulse neural network | |
| CN110929748A (en) | Motion blur image feature matching method based on deep learning | |
| CN109086663A (en) | The natural scene Method for text detection of dimension self-adaption based on convolutional neural networks | |
| CN107862680A (en) | A kind of target following optimization method based on correlation filter | |
| CN109063549A (en) | High-resolution based on deep neural network is taken photo by plane video moving object detection method | |
| CN110647909A (en) | Remote sensing image classification method based on three-dimensional dense convolution neural network | |
| CN109583377A (en) | A kind of control method, device and host computer that pipeline model is rebuild | |
| CN112001217A (en) | Multi-person human body posture estimation algorithm based on deep learning | |
| Jia et al. | Depth measurement based on a convolutional neural network and structured light | |
| Guo et al. | Decoupling semantic and edge representations for building footprint extraction from remote sensing images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| AD01 | Patent right deemed abandoned | ||
| AD01 | Patent right deemed abandoned |
Effective date of abandoning: 20221206 |