| Image Classification and face recognition are an important research direction of the computer vision analysis,and they have important application value in the fields of human computer interaction,intelligent transportation,military security and intelligent security.Sparse representation method is a newly developed signal representation method,which is based on the sparse representation of the compressed sensing theory.Because sparse representation method uses the overcomplete dictionary to decompose the signal,it has good robustness for noise and error signal.After analysing the advantages and disadvantages of the image classification of sparse representation,this paper affirms its advantages and also studies the problems existed in the image classification of sparse representation.Firstly,in the field of image representation,in order to efficiently use of the information of the original samples and virtual samles,this paper proposed a supervised sparse representation method.This method used the similarity between adjacent pixels of original image to obtain virtual image.Then,from the set composed of original samples and virtual samples,this method selected some samples which closest to the test sample as the training samples to represent the test sample and eliminated the disturbance of bad samples.In order to effectively use of the information of the pixels with moderate intensity,this paper obtained the virtual images whose pixel intensities focused on the moderate regionals by processing the pixels of original images,and then combined the original and virtual samples to represent the test sample,and greatly increased the efficiency of image information.Because the two methods all used virtual samples as the training samples,they effectively alleviated the negative effects that come from the small sample problem.Secondly,for the problem that images with different scales and noise have adverse effects on image representation,this paper proposed a mutil-scale collaborative representation method to solve this problem by combining the different scale-noise images.The multi-scale noise images have good robustness to noise problem and multi-scale problem in image classification.For the problem that samples of different classes will interfere with each other,this paper proposed a discriminant competitive representation method.This method can not only obtain an approximate representation of a test sample via all training samples,but also enable representation components generated from training samples of different classes to be discriminative and competitive,which is beneficial to correct classification of the test sample.Finally,in order to highlight the feature information of the important parts of the face and make use of the certain relation between adjacent pixels in the face images,this paper proposed a simple method to get the virtual which can highlight the facial features,and then combine the original and virtual samples to represent the test sample.This method not only makes full use of the information of different images,but also increases the number of training samples,and mitigates the impact of small sample problem.In order to make full use of different information of visible light and near-infrared face images,this paper proposed a multi-representation method based on complex matrix.In this method,the same person’s visible light and near-infrared image are combined into a complex matrix,and the image with the more important features can be given more weight,and then the complex matrix is used for recognition and classification.This method can make use of the advantages of different kinds of images to get a better recognition effect.In order to solve the problem of image classification and face detection,the problem of illumination,small sample and noise information interference is studied.Different solutions are proposed for each problem,so that image classification and face recognition have Better recognition effect. |
