Image Recognition Based On Kernel Sparse Space

With the arrival of big data age,data has the characteristics of high dimensionality,large data size and nonlinear separability.In order to solve these problems,this paper makes a deep study on the problem of image classification based on dictionary learning and kernel method.We can not only extract the internal structure of data but also reduce the data dimension using dictionary learning,sparse representation theory.We can solve the linearly inseparable problem in original data using kernel methods.The work of this thesis is stated as follows:1)Study the sparse coding theory,dictionary learning theory and kernel method theory,then the algorithm of kernel dictionary learning is described,and the algorithm complexity is analyzed on the basis of these theories.2)In order to tackle the problems of high feature dimension and linearly inseparable in original data,a recognition algorithm combining kernel dictionary learning and discriminant analysis is proposed based on dictionary learning and sparse representation.First of all,learn a kernel dictionary that explores the underlying structure of data,then obtain the sparse representations of samples by the kernel dictionary.Secondly,the linear discriminant analysis is employed to make these sparse representations more separable.Experimental results on multiple source images show that our method based on kernel dictionary learning and discriminant analysis has superior recognition performance.3)Based on kernel dictionary learning,we study the multi-kernel dictionary learning algorithm and use the multi-kernel learning algorithm instead of artificial kernel function selection.By learning the coefficients of the linear combination of the basic kernels,the kernel learning algorithm can learn the optimal combined kernel function and use it in the dictionary learning algorithm.Experimental results show that the multi-kernel dictionary learning algorithm has better recognition performance than the single-kernel dictionary learning algorithm in image classification.