CN109685076A - A kind of image-recognizing method based on SIFT and sparse coding - Google Patents
A kind of image-recognizing method based on SIFT and sparse coding Download PDFInfo
- Publication number
- CN109685076A CN109685076A CN201811481734.0A CN201811481734A CN109685076A CN 109685076 A CN109685076 A CN 109685076A CN 201811481734 A CN201811481734 A CN 201811481734A CN 109685076 A CN109685076 A CN 109685076A
- Authority
- CN
- China
- Prior art keywords
- image
- sift
- feature
- sparse coding
- algorithm
- 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]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
-
- 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
- G06V10/513—Sparse representations
Abstract
The present invention provides a kind of image-recognizing method based on SIFT and sparse coding, comprising: extracts intensive SIFT feature and pixel block feature respectively from each image of RGB-D object；According to the intensive SIFT feature extracted, SIFT feature dictionary is calculated using K-SVD algorithm, and solve the sparse coding of SIFT using match tracing MP algorithm, obtain the first characteristics of image；Using the dictionary of K-SVD study block of pixels, and the sparse coding of pixel is obtained using OMP algorithm, element characteristic is obtained by maximum pond algorithm, a plurality of element characteristics are connected as block feature, the sparse coding based on block feature is calculated, links block feature and corresponding sparse coding to obtain the second characteristics of image；All features are linked using pyramid pond algorithm, fused characteristics of image is obtained and carries out image recognition.Using image-recognizing method of the invention, image recognition accuracy rate can be improved.
Description
Technical field
The present invention relates to technical field of image processing more particularly to a kind of image recognition sides based on SIFT and sparse coding
Method.
Background technique
Object identification based on RGB-D information is a critically important project in computer vision and field of machine vision,
And there are related application such as recognition of face, gesture identification, Text region and vehicle identification.
Currently based in the object recognition algorithm of traditional algorithm, many are extracted object features using SIFT in the initial stage
In conjunction with sparse coding, recognition effect is bad.For example existing application No. is 201510567889.6 Chinese patents --- base
In the image swift nature representation method of normalization non-negative sparse coding device, and application No. is 201510874639.7 China
Patent --- the image classification method based on multi-direction contextual information and sparse coding model, can be effective based on these methods
It extracts object gradient information but the features such as color and shape can be ignored.It is directly mentioned from image with sparse coding there are also some algorithms
Take object features such as HMP algorithm, this algorithm can effectively extract the color and shape information of objects in images, but this
Algorithm can ignore object gradient information, in feature extraction and identify that upper accuracy rate is not high, cannot good identification intelligent household
Middle life familiar object.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of image-recognizing method based on SIFT and sparse coding,
Improve image recognition accuracy rate.
The present invention is implemented as follows: a kind of image-recognizing method based on SIFT and sparse coding, includes the following steps:
Step 10 extracts intensive SIFT feature and pixel block feature respectively from each image of RGB-D object；
The intensive SIFT feature that step 20, basis are extracted calculates SIFT feature dictionary using K-SVD algorithm, and uses
Match tracing MP algorithm solves the sparse coding of SIFT, obtains the first characteristics of image；
Step 30, the dictionary for learning block of pixels using K-SVD, and the sparse coding of pixel is obtained using OMP algorithm, pass through
Maximum pond algorithm obtains element characteristic, and a plurality of element characteristics are connected as block feature, calculate the sparse volume based on block feature
Code, links block feature and corresponding sparse coding to obtain the second characteristics of image；
Step 40 is respectively applied to the first characteristics of image and the second characteristics of image using pyramid pond algorithm, links institute
There is feature, obtains fused characteristics of image and carry out image recognition；
Above-mentioned steps 20 and step 30 carry out in no particular order.
Further, the extracting mode of the intensive SIFT feature in the step 10 are as follows: mentioned on the image every four pixels
An image block is taken, each image block extracts a SIFT feature；
The matrix of the intensive SIFT feature is expressed as: Y={ y1,y2,…,yu, wherein yiIt is i-th of SIFT feature, u
It is image block number.
Further, the solution procedure of the sparse coding of SIFT is as follows in the step 20:
Step 21 collects a plurality of SIFT features on every picture, uses Ys={ ys1,ys2,ys3,…,yspIndicate sample
This matrix uses Ds={ ds1,ds2,…,dsm}∈Rh×mIndicate the SIFT feature dictionary that study of Confucian classics acquistion is arrived, wherein dsiTo learn
The word or base arrived, Xs={ xs1,xs2,…xsp}∈Rp×mFor sparse coding, wherein xsRepresent ysSparse coding；
Step 22 passes through following formula Dictionary of Computing:
Wherein, | | | |FF norm is represented, | | | |0Indicate the nonzero element number in vector, o is a non-zero number, is indicated
The upper limit of sparse level；
Step 23 uses K-Means algorithm to generate initial SIFT feature dictionary in the calculating process of SIFT feature dictionary
D；
Dictionary Solve problems are decomposed into a plurality of subproblems progress substep solutions by step 24, specific using following public
Formula carries out:
Optimization dictionary is updated using SVD decomposition algorithm step by step, as follows for the update optimization process of i-th of word:
The value range that wherein i is indicated is (1,1024) for SIFT feature, is the same meaning for j and i, is all word
Word in allusion quotation, xiIndicate value of the sparse coding in the dimension of the corresponding correlated characteristic of i-th of word；
Wherein,Indicate XsRow, EiIndicate residual matrix, the word d after optimizationiWith the coefficient of optimizationIt is by svd algorithm
Applied to residual matrix EiIt obtains, using row involved in base when update, repeats this process until restraining or reaching
Preset the number of iterations；
Step 25, the SIFT feature dictionary obtained using final updated solve the sparse volume of SIFT using matching pursuit algorithm
Code, obtains the first characteristics of image.
Further, the step 30 further comprises:
Step 31, the dictionary for learning block of pixels using K-SVD；
Step 32 obtains the sparse coding of pixel using OMP algorithm；
Step 33 obtains element characteristic using maximum pond algorithm；
A plurality of element characteristics are connected as block feature by step 34, and the block feature includes color block feature, and depth block is special
It seeks peace surface normal block feature；
Step 35 calculates the sparse coding based on block feature；
Step 36 links block feature and corresponding sparse coding to obtain the second characteristics of image, second characteristics of image
PvIt indicates are as follows: Pv={ pv1,pv2,...,pv3, wherein v indicates color, depth or normal vector.
Further, the step 40 is specific as follows: gold tower basinization uses 3 × 3,2 × 2,1 × 1 three layers, is divided into altogether
14 sub-regions acquire sub-district characteristic of field HP with maximum pond algorithm in each subregionvb, b=1,2 ..., 14；
If Θv={ HPv1,HPv2,...,HPv14, wherein v indicates color, depth and surface normal, then fused object
Body image feature representation are as follows:
ψ={ Θrgb,Θdepth,Θnormal,φ}；
Later by ψ divided byThe characteristics of image finally merged is normalized, wherein ε=
0.001。
The present invention has the advantage that
1, algorithm for design frame is combined for the sparse coding and HMP algorithm of improved SIFT, and it is quasi- to improve object identification
True rate；
2, the algorithm and HMP or the sparse coding feature based on SIFT that the present invention designs can extract feature abundant, make
Gradient information is added to HMP with RGB-D information, obtains the expression with richer feature；
3, multi-channel feature integration program is proposed, by modifying HMP algorithm, to adapt to the sparse coding based on SIFT.
Detailed description of the invention
The present invention is further illustrated in conjunction with the embodiments with reference to the accompanying drawings.
Fig. 1 is a kind of image-recognizing method execution flow chart based on SIFT and sparse coding of the present invention.
Specific embodiment
As shown in Figure 1, a kind of image-recognizing method based on SIFT and sparse coding of the invention, includes the following steps:
Step 10 extracts intensive SIFT feature and pixel block feature respectively from each image of RGB-D object；
The intensive SIFT feature that step 20, basis are extracted calculates SIFT feature dictionary using K-SVD algorithm, and uses
Match tracing MP algorithm solves the sparse coding of SIFT, obtains the first characteristics of image；
Step 30, the dictionary for learning block of pixels using K-SVD, and the sparse coding of pixel is obtained using OMP algorithm, pass through
Maximum pond algorithm obtains element characteristic, and a plurality of element characteristics are connected as block feature, calculate the sparse volume based on block feature
Code, links block feature and corresponding sparse coding to obtain the second characteristics of image；
Step 40 is respectively applied to the first characteristics of image and the second characteristics of image using pyramid pond algorithm, links institute
There is feature, obtains fused characteristics of image and carry out image recognition；
Above-mentioned steps 20 and step 30 carry out in no particular order.
Preferably, the extracting mode of the intensive SIFT feature in the step 10 are as follows: on the image every four pixel extractions
One image block, each image block extract a SIFT feature；
The matrix of the intensive SIFT feature is expressed as: Y={ y1,y2,…,yu, wherein yiIt is i-th of SIFT feature, u
It is image block number.
Preferably, the solution procedure of the sparse coding of SIFT is as follows in the step 20:
Step 21 collects a plurality of SIFT features on every picture, uses Ys={ ys1,ys2,ys3,…,yspIndicate sample
This matrix uses Ds={ ds1,ds2,…,dsm}∈Rh×mIndicate the SIFT feature dictionary that study of Confucian classics acquistion is arrived, wherein dsiTo learn
The word or base arrived, Xs={ xs1,xs2,…xsp}∈Rp×mFor sparse coding, wherein xsRepresent ysSparse coding, wherein Rp×m
P and m respectively indicate the dimension of Characteristic Number with corresponding sparse coding, Rh×mH and m respectively indicate dictionary number and dictionary
Dimension, H takes 1024.
Step 22 passes through following formula Dictionary of Computing:
Wherein, | | | |FF norm is represented, | | | |0Indicate the nonzero element number in vector, o is a non-zero number, is indicated
The upper limit of sparse level；
Step 23 uses K-Means algorithm to generate initial SIFT feature dictionary in the calculating process of SIFT feature dictionary
D；
Dictionary Solve problems are decomposed into a plurality of subproblems progress substep solutions by step 24, specific using following public
Formula carries out:
Optimization dictionary is updated using SVD decomposition algorithm step by step, as follows for the update optimization process of i-th of word:
The value range that wherein i is indicated is (1,1024) for SIFT feature, is the same meaning for j and i, is all word
Word in allusion quotation, xiIndicate value of the sparse coding in the dimension of the corresponding correlated characteristic of i-th of word；
Wherein,Indicate XsRow, EiIndicate residual matrix, the word d after optimizationiWith the coefficient of optimizationIt is by svd algorithm
Applied to residual matrix EiIt obtains, using row involved in base when update, repeats this process until restraining or reaching
Preset the number of iterations；
Step 25, the SIFT feature dictionary obtained using final updated solve the sparse volume of SIFT using matching pursuit algorithm
Code, obtains the first characteristics of image.
Preferably, the step 30 further comprises:
Step 31, the dictionary for learning block of pixels using K-SVD；
Step 32 obtains the sparse coding of pixel using OMP algorithm；
Step 33 obtains element characteristic using maximum pond algorithm；
A plurality of element characteristics are connected as block feature by step 34, and the block feature includes color block feature, and depth block is special
It seeks peace surface normal block feature；
Step 35 calculates the sparse coding based on block feature
Step 36 links block feature and corresponding sparse coding to obtain the second characteristics of image, second characteristics of image
PvIt indicates are as follows: Pv={ pv1,pv2,...,pv3, v indicates color, depth and surface normal, and color point here refers to red, green
Color and blue；
Preferably, the step 40 is specific as follows: gold tower basinization uses 3 × 3,2 × 2,1 × 1 three layers, is divided into 14 altogether
Sub-regions acquire sub-district characteristic of field HP with maximum pond algorithm in each subregionvb, b=1,2 ..., 14；
If Θv={ HPv1,HPv2,...,HPv14, wherein v indicates color, depth and surface normal, then fused object
Body image feature representation are as follows:
ψ={ Θrgb,Θdepth,Θnormal,φ}；
Wherein, ΘrgbRefer to that Θ red, the combination of Θ green and Θ blue, Θ red refer to that red channel image is special
Sign, Θ green refers to green channel images feature and Θ blue refers to that blue channel characteristics of image, characteristics of image refer to
Pyramid pond algorithm is used for linking for block feature and corresponding sparse coding on each channel.
Later by ψ divided byThe characteristics of image finally merged is normalized, wherein ε=
0.001。
Since SIFT feature can extract the gradient orientation histogram of object, carrying out intensive sampling using SIFT can retain
The minutia that object is strengthened.SIFT algorithm itself has dimension rotation invariance, while having certain affine-invariant features, according to
It is indicated by the available stable image block characteristics of this characteristic.After the present invention carries out intensive sampling using SIFT feature, root
According to obtained eigenmatrix, dictionary learning is carried out using K-SVD algorithm, uses MP (matching when solving block feature later
Pursuit) algorithm obtains the rarefaction representation of block, it is therefore an objective to the minutia as much as possible for retaining image block, finally using gold
Word tower basin algorithm links all features, obtains the character representation of image.
Each image is divided into several 16*16 image block pixels, and each image block pixel is divided into 16 4*4 units.Directly
Connecing and carrying out the advantage of feature learning using sparse coding in block of pixels is that the dictionary that this block of pixels learns can save pixel
The global feature of block, including spatial relation between pixel, the geometrical characteristic for including, the connection between pixel value for including
Deng.It, can be in the hope of the sparse coding of pixel in this way by the dictionary based on block of pixels.Maximum pond algorithm is applied in unit
The available element characteristic of the corresponding sparse coding of all pixels (block of pixels that element characteristic refers to 4*4 here) then will
Element characteristic is chained up to obtain block feature.The element characteristic acquired in this way, can be with most significant feature in stick unit, and will be single
First feature, which links together, then can retain its spatial positional information while retaining significant information.For block feature, continue
It carries out feature learning and obtains block-based dictionary and sparse coding, this is to extract more abstract feature.Finally by golden word
Tower basin algorithm is respectively applied to sparse coding, block-based feature and block-based sparse coding based on SIFT, by them
It links together and can be obtained by last character representation.Wherein SIFT extracts grayscale information, directly then using sparse coding
It is to extract color, depth and normal information.
The present invention learns SIFT feature and block feature from RGB-D information, and block feature can extract face from RGB-D information
Color, shape feature, space geometry feature and direction character.Intensive SIFT feature can capture Gradient Features.Then K-SVD is used
Algorithm calculates SIFT feature dictionary.It can be obtained by more gradient informations in this way.It is calculated later using match tracing (MP)
The sparse coding of SIFT and block feature.By the splicing of block feature and corresponding sparse coding, on the basis of SIFT sparse coding,
It is compiled using the spatial pyramid pond algorithm of simple three layers segmentation 3 × 3,2 × 2,1 × 1, and using block feature and related sparse
Code feature generates final image feature, to improve image recognition accuracy rate.
In one embodiment, the object recognition algorithm of the present invention combines multiple features rarefaction representation and some classics is in China
It contains a RGB-D object data set to be compared, data set includes 51 classifications and 300 examples, and each classification includes multiple realities
Example, example are from 30 °, 45 °, and 60 ° of different angle shooting experiments have used 41788 RGB-D images, the data in an experiment
Design view, rotation, scaling, texture and the less lighting problem of texture, for Object identifying, by trained and test process,
It is carried out on data set using the method compared, there are two types of the Object identifying task of type needs to execute: Classification and Identification and reality
Example identification.Sightless object is the name in being identified by classification to determine.One example is by randomly from each classification
It is stripped out and is tested, remaining 300-51=249 object is for the training in each trace.Trained at random at 10/
Mean accuracy in test segmentation.Such as identify, the image obtained from 30 ° and 60 ° of angles be used to train, and use 45 °
The image at angle is tested.
Classification and Identification Experimental comparison results are as shown in table 1:
Table 1
| RGB | RGB-D | |
| Kernel descriptor | 80.7±2.1 | 86.5±2.1 |
| HMP | 74.7±2.5 | 82.1±3.3 |
| Improved HMP | 82.4±3.1 | 87.5±2.9 |
| VGG | 88.9±2.1 | 91.8±2.4 |
| MJSR | 91.5±1.5 | 92.8±1.3 |
Example recognition Experimental comparison results are as shown in table 2:
Table 2
| RGB | RGB-D | |
| Kernel descriptor | 90.8 | 91.2 |
| HMP | 75.8 | 78.9 |
| Improved HMP | 92.1 | 92.8 |
| MJSR | 92.6 | 92.2 |
By Experimental comparison it can be seen that object identification method of the invention is using color or by color and depth information
Combining can obtain and the competitive result of Upgraded HMP.It is whole to exceed 13 percentage points of the original algorithm of HMP
More than, either color combining information or color combining and depth information are compared with kernel function and competitive,
Color combining aspect ratio kernel function is 1.8 percentage points high, and it is 1 percentage point high to close depth than color combining.
Although specific embodiments of the present invention have been described above, those familiar with the art should be managed
Solution, we are merely exemplary described specific embodiment, rather than for the restriction to the scope of the present invention, it is familiar with this
The technical staff in field should be covered of the invention according to modification and variation equivalent made by spirit of the invention
In scope of the claimed protection.
Claims (5)
1. a kind of image-recognizing method based on SIFT and sparse coding, characterized by the following steps:
Step 10 extracts intensive SIFT feature and pixel block feature respectively from each image of RGB-D object；
The intensive SIFT feature that step 20, basis are extracted calculates SIFT feature dictionary using K-SVD algorithm, and using matching
The sparse coding that MP algorithm solves SIFT is tracked, the first characteristics of image is obtained；
Step 30, the dictionary for learning block of pixels using K-SVD, and the sparse coding of pixel is obtained using OMP algorithm, pass through maximum
Pond algorithm obtains element characteristic, and a plurality of element characteristics are connected as block feature, calculate the sparse coding based on block feature, will
Block feature and corresponding sparse coding link to obtain the second characteristics of image；
Step 40 is respectively applied to the first characteristics of image and the second characteristics of image using pyramid pond algorithm, links all spies
Sign obtains fused characteristics of image and carries out image recognition；
Above-mentioned steps 20 and step 30 carry out in no particular order.
2. a kind of image-recognizing method based on SIFT and sparse coding according to claim 1, it is characterised in that: described
The extracting mode of intensive SIFT feature in step 10 are as follows: on the image every four pixel extractions, one image block, each image
Block extracts a SIFT feature；
The matrix of the intensive SIFT feature is expressed as: Y={ y1,y2,…,yu, wherein yiIt is i-th of SIFT feature, u is figure
As block number.
3. a kind of image-recognizing method based on SIFT and sparse coding according to claim 1, it is characterised in that: described
The solution procedure of the sparse coding of SIFT is as follows in step 20:
Step 21 collects a plurality of SIFT features on every picture, uses Ys={ ys1,ys2,ys3,…,yspIndicate sample moment
Battle array, uses Ds={ ds1,ds2,…,dsm}∈Rh×mIndicate the SIFT feature dictionary that study of Confucian classics acquistion is arrived, wherein dsiStudy obtains
Word or base, Xs={ xs1,xs2,…xsp}∈Rp×mFor sparse coding, wherein xsRepresent ysSparse coding；
Step 22 passes through following formula Dictionary of Computing:
Wherein, | | | |FF norm is represented, | | | |0Indicate the nonzero element number in vector, o is a non-zero number, is indicated sparse
The horizontal upper limit；
Step 23 uses K-Means algorithm to generate initial SIFT feature dictionary D in the calculating process of SIFT feature dictionary；
Dictionary Solve problems are decomposed into a plurality of subproblems and carry out substep solutions by step 24, specifically using following formula into
Row:
Optimization dictionary is updated using SVD decomposition algorithm step by step, as follows for the update optimization process of i-th of word:
Wherein,Indicate XsRow, EiIndicate residual matrix, the word d after optimizationiWith the coefficient of optimizationIt is by svd algorithm application
In residual matrix EiIt obtains, using row involved in base when update, repeats this process until restraining or reaching default
The number of iterations；
Step 25, the SIFT feature dictionary obtained using final updated solve the sparse coding of SIFT using matching pursuit algorithm,
Obtain the first characteristics of image.
4. a kind of image-recognizing method based on SIFT and sparse coding according to claim 1, it is characterised in that: described
Step 30 further comprises:
Step 31, the dictionary for learning block of pixels using K-SVD；
Step 32 obtains the sparse coding of pixel using OMP algorithm；
Step 33 obtains element characteristic using maximum pond algorithm；
A plurality of element characteristics are connected as block feature by step 34, and the block feature includes color block feature, depth block feature and
Surface normal block feature；
Step 35 calculates the sparse coding based on block feature；
Step 36 links block feature and corresponding sparse coding to obtain the second characteristics of image, the second characteristics of image PvIt indicates
Are as follows: Pv={ pv1,pv2,...,pv3, wherein v indicates color, depth or normal vector.
5. a kind of image-recognizing method based on SIFT and sparse coding according to claim 4, it is characterised in that: described
Step 40 is specific as follows: gold tower basinization uses 3 × 3,2 × 2,1 × 1 three layers, 14 sub-regions is divided into altogether, in each sub-district
Sub-district characteristic of field HP is acquired with maximum pond algorithm on domainvb, b=1,2 ..., 14；
If Θv={ HPv1,HPv2,...,HPv14, wherein v indicates color, depth and surface normal, then fused object figure
As character representation are as follows:
ψ={ Θrgb,Θdepth,Θnormal,φ}；
Later by ψ divided byThe characteristics of image finally merged is normalized, wherein ε=0.001.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811481734.0A CN109685076A (en) | 2018-12-05 | 2018-12-05 | A kind of image-recognizing method based on SIFT and sparse coding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811481734.0A CN109685076A (en) | 2018-12-05 | 2018-12-05 | A kind of image-recognizing method based on SIFT and sparse coding |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109685076A true CN109685076A (en) | 2019-04-26 |
Family
ID=66187281
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811481734.0A Pending CN109685076A (en) | 2018-12-05 | 2018-12-05 | A kind of image-recognizing method based on SIFT and sparse coding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109685076A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210311A (en) * | 2019-04-30 | 2019-09-06 | 杰创智能科技股份有限公司 | A kind of face identification method based on channel characteristics fusion rarefaction representation |
| CN111126513A (en) * | 2020-01-20 | 2020-05-08 | 柳州智视科技有限公司 | Universal object real-time learning and recognition system and learning and recognition method thereof |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100124377A1 (en) * | 2008-11-19 | 2010-05-20 | Nec Laboratories America, Inc. | Linear spatial pyramid matching using sparse coding |
| CN108090504A (en) * | 2017-12-04 | 2018-05-29 | 泉州装备制造研究所 | Object identification method based on multichannel dictionary |
| CN108090505A (en) * | 2017-12-04 | 2018-05-29 | 泉州装备制造研究所 | A kind of object identification method based on multichannel dictionary |
-
2018
- 2018-12-05 CN CN201811481734.0A patent/CN109685076A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100124377A1 (en) * | 2008-11-19 | 2010-05-20 | Nec Laboratories America, Inc. | Linear spatial pyramid matching using sparse coding |
| CN108090504A (en) * | 2017-12-04 | 2018-05-29 | 泉州装备制造研究所 | Object identification method based on multichannel dictionary |
| CN108090505A (en) * | 2017-12-04 | 2018-05-29 | 泉州装备制造研究所 | A kind of object identification method based on multichannel dictionary |
Non-Patent Citations (2)
| Title |
|---|
| 兰晓东: "基于RGB-D视觉信息的物体识别算法研究", 《中国优秀硕士学位论文全文数据库》 * |
| 卢良锋等: "基于RGB特征与深度特征融合的物体识别算法", 《计算机工程》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210311A (en) * | 2019-04-30 | 2019-09-06 | 杰创智能科技股份有限公司 | A kind of face identification method based on channel characteristics fusion rarefaction representation |
| CN110210311B (en) * | 2019-04-30 | 2021-05-04 | 杰创智能科技股份有限公司 | Face recognition method based on channel feature fusion sparse representation |
| CN111126513A (en) * | 2020-01-20 | 2020-05-08 | 柳州智视科技有限公司 | Universal object real-time learning and recognition system and learning and recognition method thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110532900B (en) | Facial expression recognition method based on U-Net and LS-CNN | |
| Rekha et al. | Shape, texture and local movement hand gesture features for indian sign language recognition | |
| Larios et al. | Automated insect identification through concatenated histograms of local appearance features: feature vector generation and region detection for deformable objects | |
| CN105488536B (en) | A kind of agricultural pests image-recognizing method based on multiple features depth learning technology | |
| CN105205449B (en) | Sign Language Recognition Method based on deep learning | |
| CN107886062B (en) | Image processing method, system and server | |
| CN102521616B (en) | Pedestrian detection method on basis of sparse representation | |
| CN103544504B (en) | Scene character recognition method based on multi-scale map matching core | |
| CN104036293B (en) | Rapid binary encoding based high resolution remote sensing image scene classification method | |
| CN108154133B (en) | Face portrait-photo recognition method based on asymmetric joint learning | |
| CN107862680B (en) | Target tracking optimization method based on correlation filter | |
| Jain et al. | Random walk-based feature learning for micro-expression recognition | |
| Kobchaisawat et al. | Thai text localization in natural scene images using convolutional neural network | |
| Hassan et al. | Pose invariant palm vein identification system using convolutional neural network | |
| Gonçalves et al. | Carcass image segmentation using CNN-based methods | |
| CN108491863A (en) | Color image processing method based on Non-negative Matrix Factorization and convolutional neural networks | |
| CN113011243A (en) | Facial expression analysis method based on capsule network | |
| CN109685076A (en) | A kind of image-recognizing method based on SIFT and sparse coding | |
| Morioka et al. | Learning Directional Local Pairwise Bases with Sparse Coding. | |
| KR20130059212A (en) | Robust face recognition method through statistical learning of local features | |
| CN105868711A (en) | Method for identifying human body behaviors based on sparse and low rank | |
| CN105844605B (en) | Based on the human face portrait synthetic method adaptively indicated | |
| Yin et al. | Object recognition in art drawings: Transfer of a neural network | |
| Cintas et al. | Automatic ear detection and segmentation over partially occluded profile face images | |
| Risojević et al. | Unsupervised learning of quaternion features for image classification |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190426 |
|
| WD01 | Invention patent application deemed withdrawn after publication |