CN103258210B - A kind of high-definition image classification method based on dictionary learning - Google Patents

Abstract

The invention discloses a kind of high-definition image classification method based on dictionary learning, relate to digital image processing field.The method comprise the steps that the visual signature extracting all high-definition image samples；For described visual signature, high-definition image sample is carried out sparse coding, constantly carried out dictionary learning by alternative manner, until error in classification is less than threshold value, determines the classifying dictionary of high-definition image classification and according to visual signature, the disturbance degree of reconstructed error is determined the weights of correspondence；The weights corresponding with visual signature according to the dictionary of described high-definition image classification set up image non-linear grader, determine the classification belonging to described high-definition image.The present invention can carry out dictionary learning by sparse coding, obtain the sparse coding with higher discrimination, so that the sample space of high-definition image data set is distributed by sorting technique has stronger adaptivity, complicated image is had more preferable robustness, there is the strongest versatility and higher practical value.

Description

A kind of high-definition image classification method based on dictionary learning

Technical field

The present invention relates to digital image processing field, be specifically related to a kind of high-definition image classification side based on dictionary learning Method.

Background technology

At present, along with the developing rapidly of computer network, digital media technology and the development of intelligent information processing technology and Extensively application, large-scale image resource constantly occurs.Facing to the image information of magnanimity, how image classified or mark Quickly, efficiently to retrieve from mass image data in interested image artificial intelligence and pattern recognition Study hotspot, have in fields such as scientific research, national defense and military, commercial production, Aero-Space, biomedicine, traffic monitorings It is widely applied.

History decades of research to image classification, period emerged in large numbers various sorting technique based on different theories, but It is that this field never forms a unified theoretical system and instructs design and the realization of new sorting technique.In recent years, The common method of image classification includes statistical method and structural approach, but statistical method shows active in image classification field, Create the newest method, such as neural net method, support vector machine (Support Vector Machine, SVM) method And Boosting method.According to the difference of implementation method, statistical method can be divided into:

1) unsupervised classification method: also known as clustering methodology, carry out category division according to the similarity between pattern, by similar The mode division that property is strong is same classification.The method need not there is categorical data more deep understanding, is formed adaptively Categorized data set, but classifying quality is not ideal in complex data.

2) supervised classification: according to the training sample of previously known classification, obtains all kinds of distribution scale at feature space, And utilize this regularity of distribution that unknown data is carried out sorting technique.The method makes full use of the priori of categorical data, and Can pass through repeated examinations training sample, improve nicety of grading, therefore the method is extensively applied in high-definition image is classified.

Support vector machine (SVM) is widely used in the classifying quality that image classification domain variability acquirement is good at present, the earliest Support vector machine (SVM) method is at " Support-Vector Networks, Machine by Vapnik and Chervonenkis Learning, 20,1995. " literary composition proposes.Support vector machine (SVM) is new machine learning method, has complete statistics Theory of learning basis, it uses structural risk minimization to replace the empiric risk based on large sample in traditional statistics Littleization principle, overcoming neutral net is affected greatly by the complicated network structure and sample size, easily occurred study or The deficiency of low generalization ability, has outstanding learning capacity and Generalization Ability for Small Sample Database analysis, in pattern recognition and Function Estimation obtains effective application, but the method still suffered from problems with:

(1) large-scale training sample is difficult to carry out by SVM algorithm, due to SVM be made by quadratic programming solve support to Amount, and solve quadratic programming and will relate to the calculating (m is the number of sample) of m rank matrix, the storage of this matrix when m large number Substantial amounts of machine internal memory and operation time will be expended with calculating.

(2) complexity of SVM method quickly can increase with the increase of Characteristic Number, is used for training grader and test knot The sample size of fruit can increase along with the quantity exponentially relation of feature；If increase the strongest feature of adaptability or with existing spy Levy the feature of strong correlation, the classification capacity of grader can be made on the contrary to decline, reduce the ability of system identification classification.

Later, France scholar was at I.Guyon document " Guyon I, Weston J, Barnhill, et al.Gene Selection for Cancer Classification Using Support Vector Machines.Machine Learning, 2002,46 (1/2/3): 389-422 " propose this algorithm in the middle of and think that use SVM-RFE algorithm can ensure that spy During levying sequence, reservation optimization character subset is when being ranked up feature, and the method uses the differentiation letter of support vector machine Information in number realizes.SVM-RFE method is the process of a circulation, comprises the steps of 1) train with current data set Grader, obtains the relevant information of used feature according to gained grader；2) according to the rule formulated in advance, all spies are calculated The ranking criteria mark levied；3) feature removed corresponding to minimum ranking criteria mark is concentrated in current data.This cyclic process Going to terminate algorithm when remaining last variable in feature set to perform result is the string spy according to feature importance ranking Levy sequence number list.

The present invention program is based on the deficiency existing for above-mentioned sorting technique, it is proposed that Nonlinear Classification based on dictionary learning Method, utilizes dictionary learning method adaptive training to go out the potential general character of image pattern in classification, efficiently differentiates different figure The characteristic of picture, and weights can be given automatically according to the dependency of feature in dictionary learning process, improve the spy of strong correlation Levy the impact on classification capacity, be effectively improved the efficiency of classification.

Summary of the invention

It is an object of the invention to provide high-definition image a kind of sorting technique, the method can solve the problem that current supervised classification side The classification speed of method is slow, complexity increases and exponential increase, feature that dependency is the strongest affect nicety of grading along with Characteristic Number Problem.

The present invention provides a kind of high-definition image classification method based on dictionary learning, comprises the steps:

S1: extract the visual signature of all high-definition image samples；

S2: for described visual signature, high-definition image sample is carried out sparse coding, constantly carried out by alternative manner Dictionary learning, until error in classification is less than threshold value, determines the classifying dictionary of high-definition image classification and according to visual signature to reconstruct The disturbance degree of error determines the weights of correspondence；

S3: set up image non-linear classification according to the weights that the dictionary of described high-definition image classification is corresponding with visual signature Device, determines the classification belonging to described high-definition image.

Above-mentioned high-definition image classification method based on dictionary learning, wherein the extraction all high-definition images sample described in step S1 This visual signature, is carried out as follows:

S1.1: extract feature (color, texture, shape, histograms of oriented gradients (HOG), the word bag feature of every image (BoW), scale invariant feature conversion (SIFT) etc.) x_i, i=1 ..., k, k are the quantity of visual signature；

S1.2: the visual signature obtaining image according to S1.1 is X=[x₁,…,x_k]。

Above-mentioned high-definition image classification method based on dictionary learning, wherein special for described vision described in step S2 Levy, high-definition image sample is carried out sparse coding, constantly carried out dictionary learning by alternative manner, until error in classification is less than threshold Value, determines the classifying dictionary of high-definition image classification and the disturbance degree of reconstructed error determines according to visual signature the weights of correspondence, Carry out as follows:

S2.1: initialize the classifying dictionary D of high-definition image all categories^p,0Weights ω with visual signature^p,0；

For the high-definition image sample of each classification, initializing dictionary is

The weights of visual signature are

${\mathrm{\ω}}^{p,0}={\left\{{\mathrm{\ω}}_i\right\}}_1^m,{\mathrm{\ω}}_i=1$

Wherein, x_iFor the characteristic vector of high-definition image, m is characterized value number.

S2.2: high-definition image sample is carried out rarefaction representation, obtains the corresponding sparse coding of image pattern；

For each classification sample of high-definition image, its sparse coding matrix For the sparse coding of the jth image pattern of pth classification, C is the classification number of high-definition image sample, m_pFor pth classification Sample number, t is iterations, then sparse coding matrix A should meet

$\underset{{\{D^{p,t},A^{p,t}\}}_{p=\mathrm{1...}C}}{\min }\underset{p=1}{\overset{C}{\Σ}}\{{||X^p-D^{p,t}\·A^{p,t}||}_2^2+\mathrm{\λ}\underset{i=1}{\overset{m_p}{\Σ}}||{\mathrm{\α}}_i^{p,t}||\}$

Wherein, λ is balance parameters.

S2.3: calculate the reconstructed error of each classification in high-definition image sample；

For each classification sample of high-definition image, its reconstructed error is

$R^{p,t}\left(X^p\right)=\underset{i=1}{\overset{n}{\Σ}}\left|x_i^{p,t}-D^{p,t}\·{\mathrm{\α}}_i^{p,t}\right|$

Wherein

S2.4: utilize Nonlinear Programming Theory that the reconstructed error of all categories is optimized, determine the weight of visual signature Want degree；

For the visual signature of high-definition image sample, its significance level matrix isAccording to regarding Feel that feature is the most important, the principle that reconstructed error is the fewest, characteristic importance β^p,tShould meet

$\underset{{\left\{{\mathrm{\β}}^{p,t}\right\}}_{p=\mathrm{1...}C}}{\min }\underset{p=1}{\overset{C}{\Σ}}({\left({\mathrm{\β}}^{p,t}\right)}^T\·R^{p,t}(X^p)-{\left({\mathrm{\β}}^{c,t}\right)}^T\·R^{c,t}(X^p\left)\right)$

$\begin{array}{cc}s.t.& 0\&le;{\mathrm{\&beta;}}_j^{p,t}<1,\underset{j=1}{\overset{m}{\&Sigma;}}{\mathrm{\&beta;}}_j^{p,t}=1\end{array}$

$c=\arg \underset{\{q,q\&NotEqual;p\}}{\min }{R}^{q,t}\left(X^p\right)$

S2.5: utilize characteristic importance obtained above to update weights and the classifying dictionary of visual signature；

For obtained visual signature and importance degree β^p,t, update the weights of visual signature

${\mathrm{\&omega;}}_j^{p,t}=\left\{\begin{array}{c}{\mathrm{\&beta;}}_j^{p,t},{\mathrm{\&beta;}}_j^{p,t}<\mathrm{\&mu;}\\ 1,{\mathrm{\&beta;}}_j^{p,t}\&GreaterEqual;\mathrm{\&mu;}\end{array}\right.$

ω^p,t=ω^p,t-1·ω^p,t

And classifying dictionary

D^p,t=D^p,t-1·ω^p,t-1

S2.6: judge that the reconstructed error summation of all categories, whether less than threshold epsilon, if being not less than threshold value, then returns step Rapid S2.2, until meeting reconstructed error less than threshold condition be.

Above-mentioned high-definition image classification method based on dictionary learning, wherein described in step S3 according to described high-definition image The weights that the dictionary of classification is corresponding with visual signature set up image non-linear grader, determine the class belonging to described high-definition image Not, carry out as follows:

According to the weight matrix W of described sparse coding matrix A, classifying dictionary D and visual signature, build high-definition image non- Linear classifier is:

$c^{*}=\arg \underset{p}{\min }{({\mathrm{\&omega;}}^{p,t^{*}-1}\&CenterDot;{\mathrm{\&beta;}}^{p,t^{*}})}^T\left|y-D^{p,t^{*}}\&CenterDot;{\mathrm{\&alpha;}}^{p,t^{*}}\right|$

Wherein, y is the visual feature vector of high-definition image sample to be sorted, t^*For final iterations,For spy Levy importance degree, D^p,tFor classifying dictionary.

Technical scheme compared with prior art has the advantage that

(1) due to the fact that employing dictionary learning method carries out sparse volume to generic image with identical classifying dictionary Code, has taken into full account association potential between similar image, has been effectively improved the precision of classification；

(2) due to the fact that and carry out giving power to each visual characteristic to the disturbance degree of reconstructed error according to visual signature Value, solve adaptability the strongest or have the feature reaction result to nicety of grading of strong correlation with existing feature, Further increase the precision of classification；

(3) emulation experiment shows, the nicety of grading of the more existing supervised classification method of the present invention is higher with classification effectiveness relatively Good.

Accompanying drawing explanation

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing In having technology to describe, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is only this Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to Other accompanying drawing is obtained according to these accompanying drawings.

Fig. 1 is the high-definition image classification method flow chart based on dictionary learning in the embodiment of the present invention；

Fig. 2 is the present invention with conventional images sorting technique NS, SVM-RFE, LP-Boost at standard drawing image set the 17 The contrast and experiment figure of the different visual signatures of category Oxford Flowers data set；

Fig. 3 is the present invention with existing image classification method NS, SVM-RFE, LP-Boost at standard drawing image set the 102 The contrast and experiment figure of the different visual signatures on category Oxford Flowers data set；

Fig. 4 is the present invention with conventional images sorting technique NS, SVM-RFE, LP-Boost at standard drawing image set the The contrast and experiment figure of the different visual signatures on Caltech 101 dataset.

Detailed description of the invention

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on Embodiment in the present invention, those of ordinary skill in the art obtained under not making creative work premise all other Embodiment, broadly falls into the scope of protection of the invention.

With reference to Fig. 1, the high-definition image classification method based on dictionary learning of the technical program comprises the following steps:

Step 1: extract the visual signature of all high-definition image samples.

(1) feature (color, texture, shape, histograms of oriented gradients (HOG), the word bag feature of every image are extracted (BoW), scale invariant feature conversion (SIFT) etc.) x_i, i=1 ..., k, k are the quantity of visual signature；

(2) basis (1) obtains the visual signature of image is X=[x₁,…,x_k]。

Step 2: for described visual signature, carries out sparse coding to high-definition image sample, continuous by alternative manner Carry out dictionary learning, until error in classification is less than threshold value, determine the classifying dictionary of high-definition image classification and according to visual signature pair The disturbance degree of reconstructed error determines the weights of correspondence

(1) the classifying dictionary D of high-definition image all categories is initialized^p,0Weights ω with visual signature^p,0；

For the high-definition image sample of each classification, initializing dictionary is

The weights of visual signature are

${\mathrm{\&omega;}}^{p,0}={\left\{{\mathrm{\&omega;}}_i\right\}}_1^m,{\mathrm{\&omega;}}_i=1$

Wherein, xi is the characteristic vector of high-definition image, and m is characterized value number.

(2) high-definition image sample is carried out rarefaction representation, obtain the corresponding sparse coding of image pattern；

For each classification sample of high-definition image, its sparse coding matrix For the sparse coding of the jth image pattern of pth classification, C is the classification number of high-definition image sample, m_pFor pth classification Sample number, t is iterations, then sparse coding matrix A should meet

$\underset{{\{D^{p,t},A^{p,t}\}}_{p=\mathrm{1...}C}}{\min }\underset{p=1}{\overset{C}{\&Sigma;}}\{{||X^p-D^{p,t}\&CenterDot;A^{p,t}||}_2^2+\mathrm{\&lambda;}\underset{i=1}{\overset{m_p}{\&Sigma;}}||{\mathrm{\&alpha;}}_i^{p,t}||\}$

Wherein, λ is balance parameters.

(3) reconstructed error of each classification in high-definition image sample is calculated；

For each classification sample of high-definition image, its reconstructed error is

$R^{p,t}\left(X^p\right)=\underset{i=1}{\overset{n}{\&Sigma;}}\left|x_i^{p,t}-D^{p,t}\&CenterDot;{\mathrm{\&alpha;}}_i^{p,t}\right|$

Wherein

(4) utilize Nonlinear Programming Theory that the reconstructed error of all categories is optimized, determine the important of visual signature Degree；

For the visual signature of high-definition image sample, its significance level matrix isAccording to regarding Feel that feature is the most important, the principle that reconstructed error is the fewest, characteristic importance β^p,tShould meet

$\underset{{\left\{{\mathrm{\&beta;}}^{p,t}\right\}}_{p=\mathrm{1...}C}}{\min }\underset{p=1}{\overset{C}{\&Sigma;}}({\left({\mathrm{\&beta;}}^{p,t}\right)}^T\&CenterDot;R^{p,t}(X^p)-{\left({\mathrm{\&beta;}}^{c,t}\right)}^T\&CenterDot;R^{c,t}(X^p\left)\right)$

$\begin{array}{cc}s.t.& 0\&le;{\mathrm{\&beta;}}_j^{p,t}<1,\underset{j=1}{\overset{m}{\&Sigma;}}{\mathrm{\&beta;}}_j^{p,t}=1\end{array}$

$c=\arg \underset{\{q,q\&NotEqual;p\}}{\min }{R}^{q,t}\left(X^p\right)$

(5) utilize characteristic importance obtained above to update weights and the classifying dictionary of visual signature；

For obtained visual signature and importance degree β^p,t, update the weights of visual signature

${\mathrm{\&omega;}}_j^{p,t}=\left\{\begin{array}{c}{\mathrm{\&beta;}}_j^{p,t},{\mathrm{\&beta;}}_j^{p,t}<\mathrm{\&mu;}\\ 1,{\mathrm{\&beta;}}_j^{p,t}\&GreaterEqual;\mathrm{\&mu;}\end{array}\right.$

ω^p,t=ω^p,t-1·ω^p,t

And classifying dictionary

D^p,t=D^p,t-1·ω^p,t-1

Wherein,It is characterized importance degree,Weights for the visual signature after updating.

(6) judge that the reconstructed error summation of all categories, whether less than threshold epsilon, if being not less than threshold value, then returns step (2), until meeting reconstructed error less than threshold condition and being.

Step 3: the weights corresponding with visual signature according to the dictionary of described high-definition image classification set up image non-linear Grader, determines the classification belonging to described high-definition image.

According to the weight matrix W of described sparse coding matrix A, classifying dictionary D and visual signature, build high-definition image non- Linear classifier is:

$c^{*}=\arg \underset{p}{\min }{({\mathrm{\&omega;}}^{p,t^{*}-1}\&CenterDot;{\mathrm{\&beta;}}^{p,t^{*}})}^T\left|y-D^{p,t^{*}}\&CenterDot;{\mathrm{\&alpha;}}^{p,t^{*}}\right|$

Wherein, y is the visual feature vector of high-definition image sample to be sorted, t^*For final iterations,For spy Levy importance degree, D^p,tFor classifying dictionary.

Effectiveness and the practicality of the inventive method is verified below by way of emulation experiment.

Emulation content:

(1) use the form of contrast experiment, select two representative sorting techniques enterprising at same image set Row test, to verify effectiveness of the invention.Specifically chosen is nearest subspace (NS) method proposed by Li et al., specifically List of references " Li, S.Z.:Face recognition based on nearest linear combinations.In: Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition,pp.839–844.IEEE Computer Society,Washington,DC,USA (1998) the SVM-RFE method ", by I.Guyon, J.Westo and Barnhill proposed, particular reference " Guyon I, Weston J,Barnhill,et al.Gene Selection for Cancer Classification Using Support Vector Machines.Machine Learning, 2002,46 (1/2/3): 389-422. " and by P.Gehler The LP-Boost method proposed with S.Nowozin, particular reference " P.Gehler and S.Nowozin.On feature combination for multiclass object classification.In International Conference on Computer Vision.IEEE,2009.”

(2) standard drawing image set the 17category Oxford Flowers data set, the 102 are used Category Oxford Flowers data set, the Caltech 101 dataset carry out emulation experiment, to verify this The invention classifying quality to different images collection, concrete simulated conditions refers to the description of each experiment.

Experiment one: image set the 17 category Oxford Flowers data set is the flowers that Britain is common Image data set, it includes 17 class flowers, and each class comprises 80 width images, and this experiment is respectively in single feature situation, multiple In feature situation, the method to the present invention carries out accuracy comparison.In the case of single feature, this experiment is respectively in color (Color), to NS, SVM-in the feature such as histograms of oriented gradients (HOG), word bag feature (BoW) and all features extracted RFE, LP-Boost and the inventive method carry out nicety of grading contrast, and its classification results is as in figure 2 it is shown, simulation result shows: this Inventive method nicety of grading under different characteristic is all high than NS, SVM-RFE, LP-Boost method.

Experiment two: image set the 102 category Oxford Flowers data set is the flowers that Britain is common Image data set, it includes 102 class flowers, and each class comprises 40 to 258 width images, and this experiment is respectively in single feature feelings In condition, multiple feature situation, the method to the present invention carries out accuracy comparison.In the case of single feature, this experiment is respectively in color (Color), to NS, SVM-RFE, LP-Boost and Ben Fa in the feature such as histograms of oriented gradients (HOG), word bag feature (BoW) Bright method carries out nicety of grading contrast, and its classification results is as it is shown on figure 3, simulation result shows: the inventive method is in different characteristic Under nicety of grading all high than NS and SVM method.

Experiment three: image set the Caltech 101 dataset is that California Institute of Technology is for testing recognizer Image data set, it includes 101 class images, and each class comprises 40 to 800 width images, and the size of image is about 300 × 200 pictures Element, this experiment carries out accuracy comparison to the method for the present invention respectively in single feature situation, multiple feature situation.Single spy In the case of levying, this experiment is right in the features such as color (Color), histograms of oriented gradients (HOG), word bag feature (BoW) respectively NS, SVM-RFE, LP-Boost and the inventive method carry out nicety of grading contrast, its classification results as shown in Figure 4, simulation result Show: the inventive method nicety of grading under different characteristic is all high than NS, SVM-RFE, LP-Boost method.

Test result indicate that, the nicety of grading of the method for the present invention either combines feelings in single feature or multiple features Under condition, all ratios are higher than existing supervised classification method.

The high-definition image classification method based on dictionary learning provided the embodiment of the present invention above, has carried out detailed Jie Continuing, principle and the embodiment of the present invention are set forth by specific case used herein, and the explanation of above example is only It is the method and core concept thereof being adapted to assist in and understanding the present invention；Simultaneously for one of ordinary skill in the art, according to this Bright thought, the most all will change, and in sum, this specification content should not be managed Solve as limitation of the present invention.

Claims (8)

1. a high-definition image classification method based on dictionary learning, it is characterised in that comprise the steps:

Extract the color of every image, texture, shape, histograms of oriented gradients, word bag feature, scale invariant feature converting characteristic As its visual signature X=[x₁,…,x_k], k is the quantity of visual signature；

For described visual signature, high-definition image sample is carried out sparse coding, constantly carried out dictionary by alternative manner Practise, until error in classification is less than threshold value, determine the classifying dictionary of high-definition image classification and according to visual signature to reconstructed error Disturbance degree determines the weights of correspondence；

The weights corresponding with visual signature according to the dictionary of described high-definition image classification set up image non-linear grader, determine Classification belonging to described high-definition image；

Wherein, high-definition image sample being carried out rarefaction representation, the process obtaining the corresponding sparse coding of image pattern includes:

For each classification sample of high-definition image, its sparse coding matrix For the sparse coding of the jth image pattern of pth classification, C is the classification number of high-definition image sample, m_pFor pth classification Sample number, t is iterations, then sparse coding matrix A should meet

$\underset{{\{D^{p,t},A^{p,t}\}}_{p=\mathrm{1...}C}}{\&Sigma;}\underset{p=1}{\overset{C}{\&Sigma;}}\{{||X^p-D^{p,t}\&CenterDot;A^{p,t}||}_2^2+\mathrm{\&lambda;}\underset{i=1}{\overset{m_p}{\&Sigma;}}||{\mathrm{\&alpha;}}_i^{p,t}||\}$

Wherein, D^p,tFor classifying dictionary, λ is balance parameters, X^pFor initializing dictionary.

2. high-definition image classification method based on dictionary learning as claimed in claim 1, it is characterised in that described determine high definition The classifying dictionary of image category and the disturbance degree of reconstructed error is determined that the weights of correspondence include according to visual signature:

Initialize the classifying dictionary D of high-definition image all categories^p,0Weights ω with visual signature^p,0；

For initialized classifying dictionary D^p,0, use iterative method constantly classifying dictionary to be learnt so that error in classification is less than Threshold epsilon, obtains final classifying dictionary and visual signature weights.

3. high-definition image classification method based on dictionary learning as claimed in claim 1, it is characterised in that described initialization is high The classifying dictionary D of clear image all categories^p,0Weights ω with visual signature^p,0Specifically include:

For the high-definition image sample of each classification, initializing dictionary is

The weights of visual signature are

${\mathrm{\&omega;}}^{p,0}={\left\{{\mathrm{\&omega;}}_i\right\}}_1^m,{\mathrm{\&omega;}}_i=1$

Wherein, x_iFor the characteristic vector of high-definition image, m is characterized value number, and n is natural number.

4. high-definition image classification method based on dictionary learning as claimed in claim 1, it is characterised in that described for initially The classifying dictionary D changed^p _, ⁰, use iterative method constantly classifying dictionary to be learnt so that error in classification is less than threshold epsilon, obtains Whole classifying dictionary and visual signature weights also include:

High-definition image sample is carried out rarefaction representation, obtains the corresponding sparse coding of image pattern；

Calculate the reconstructed error of each classification in high-definition image sample；

Utilize Nonlinear Programming Theory that the reconstructed error of all categories is optimized, determine the significance level of visual signature；

Utilize characteristic importance obtained above to update weights and the classifying dictionary of visual signature；

Judge that the reconstructed error summation of all categories, whether less than threshold epsilon, if not less than threshold value, then returns step to high definition figure Decent carries out rarefaction representation, obtains the corresponding sparse coding of image pattern, until meeting reconstructed error less than threshold condition is Only.

5. high-definition image classification method based on dictionary learning as claimed in claim 4, it is characterised in that described calculating high definition In image pattern, the reconstructed error of each classification includes:

For each classification sample of high-definition image, its reconstructed error is

$R^{p,t}\left(X^p\right)=\underset{i=1}{\overset{n}{\&Sigma;}}\left|x_i^{p,t}-D^{p,t}\&CenterDot;{\mathrm{\&alpha;}}_i^{p,t}\right|$

WhereinN is natural number, D^p _, ^tFor classifying dictionary,Jth image pattern for pth classification Sparse coding.

6. high-definition image classification method based on dictionary learning as claimed in claim 4, it is characterised in that described utilize non-thread The reconstructed error of all categories is optimized by property planning theory, determines that the significance level of visual signature includes:

For the visual signature of high-definition image sample, its significance level matrix isAccording to visual signature The most important, that reconstructed error is the fewest principle, characteristic importance β^p _, ^tShould meet

$\underset{{\left\{{\mathrm{\&beta;}}^{p,t}\right\}}_{p=\mathrm{1...}C}}{\min }\underset{p=1}{\overset{C}{\&Sigma;}}({\left({\mathrm{\&beta;}}^{p,t}\right)}^T\&CenterDot;R^{p,t}(X^p)-{\left({\mathrm{\&beta;}}^{c,t}\right)}^T\&CenterDot;R^{c,t}(X^p\left)\right)$

$\begin{array}{cc}s.t.& 0\&le;{\mathrm{\&beta;}}_j^{p,t}<1,\underset{j=1}{\overset{m}{\&Sigma;}}{\mathrm{\&beta;}}_j^{p,t}=1\end{array}$

It is characterized importance degree.

7. high-definition image classification method based on dictionary learning as claimed in claim 4, it is characterised in that described utilize above-mentioned The characteristic importance obtained includes to the weights and classifying dictionary updating visual signature:

For obtained visual signature and importance degree β^p _, ^t, update the weights of visual signature

${\mathrm{\&omega;}}_j^{p,i}=\left\{\begin{array}{c}{\mathrm{\&beta;}}_j^{p,i},{\mathrm{\&beta;}}_j^{p,i}<\mathrm{\&mu;}\\ 1,{\mathrm{\&beta;}}_j^{p,i}\&GreaterEqual;\mathrm{\&mu;}\end{array}\right.$

ω^p _, ^t=ω^p _, ^t-1·ω^p _, ^t

And classifying dictionary

D^p _, ^t=D^p _, ^t-1·ω^p _, ^t-1

Wherein,It is characterized importance degree,Weights for the visual signature after updating.

8. high-definition image classification method based on dictionary learning as claimed in claim 1, it is characterised in that according to described height The weights that the dictionary of clear image category is corresponding with visual signature set up image non-linear grader, determine belonging to described high-definition image Classification, also include:

According to the weight matrix W of described sparse coding matrix A, classifying dictionary D and visual signature, build high-definition image non-linear Grader is:

$c^{*}=\arg \underset{p}{\min }{({\mathrm{\&omega;}}^{p,t^{*}-1}\&CenterDot;{\mathrm{\&beta;}}^{p,t^{*}})}^T\left|y-D^{p,t^{*}}\&CenterDot;{\mathrm{\&alpha;}}^{p,t^{*}}\right|$

Wherein, y is the visual feature vector of high-definition image sample to be sorted, t^*For final iterations,It is characterized weight Spend, D^p _, ^tFor classifying dictionary,Weights for the visual signature after updating.