An Image Recognition Method Based on Sparse Representation

Abstract

The invention discloses an image recognition method based on sparse representation. According to the invention, training samples can be adaptively selected for multiple rounds of training; a pluralityof dictionaries are learned, each dictionary learns other samples with poor dictionary representation precision in a targeted mode, each dictionary corresponds to one targeted weak classifier, classification results of the weak classifiers are subjected to weighted combination, and the recognition precision of traditional sparse representation applied to classification problems is improved. The method comprises the following steps: (1) learning a plurality of dictionaries and corresponding weak classifiers based on an adaptive enhanced dictionary learning process, and calculating classifier weight coefficients; and (2) calculating a sparse representation vector of the to-be-classified data based on the plurality of dictionaries learned in the step (1), performing classification by using corresponding weak classifiers, and performing weighted combination on identification results of the weak classifiers to obtain a final identification result.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an image recognition method based on sparse representation. Background technique [0002] In recent years, the sparse representation algorithm has been widely used as a tool for image classification. It learns dictionaries through training or uses training samples directly as dictionaries, performs sparse coding on test data based on dictionaries, and achieves classification by comparing reconstruction errors of samples on category dictionaries. In order to improve the performance of dictionary methods on classification problems, existing methods improve on the traditional sparse representation method KSVD, adding category information to the objective function, so as to achieve the purpose of constraining the dictionary. For example, in DKSVD (Discriminative KSVD), on the basis of KSVD, a classification error term composed of a sample label matrix, a classification matri...

AI Technical Summary

Problems solved by technology

However, the complexity of different samples is different, so all samples contribute the same to the dictionary training, on the one hand, it may bring information redundancy, on the other hand, it may lead to the lack of information of complex samples

Images