A Learning Method for Multilayer Discriminative Convolutional Sparse Coding

Abstract

The invention discloses a multi-layer discriminative convolutional sparse coding learning method, which includes embedding a discriminative information module and constructing a multi-layer discriminative convolutional sparse coding module; Flow type constraints or classification error items make the feature representation have better discrimination ability; the construction of multi-layer discriminative convolutional sparse coding module refers to the structural framework of convolutional neural network, and is based on the research basis of deep discriminant analysis-comprehensive dictionary learning method Above, try to extend the discriminative convolutional sparse coding model of single layer to multi-layer, and build a multi-layer discriminative convolutional sparse coding model. The combined use of the convolutional sparse coding module makes the features calculated by the convolutional sparse coding method not lack of discrimination ability, which ensures its application in tasks such as image classification and face recognition.

Description

technical field [0001] The invention relates to the technical field of convolutional sparse coding, in particular to a learning method for multi-layer discrimination convolutional sparse coding. Background technique [0002] The traditional sparse representation method needs to pull the image into a column vector when extracting features, which causes the loss of spatial information and structural information between image pixels. Compared with traditional sparse coding methods, dictionary atoms in convolutional sparse coding have better structural and spatial invariance. Since one dictionary atom in convolutional sparse coding can globally represent all edges in a specific direction, it has stronger representation power. [0003] Considering the geometric structure information of the image, A.Szlam et al. proposed a convolutional matching pursuit (CMP) algorithm. The algorithm treats an image as a linear approximation of a set of sparse feature responses summed with a set...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the deficiencies of the prior art, the present invention provides a multi-layer discriminative convolutional sparse coding learning method, which solves the difficult problems raised in the background technology

Images