Image Feature Extraction Method Based on Codebook Block Sparse Non-negative Sparse Coding

Abstract

The present invention is based on the image feature extraction method of code book block sparse non-negative sparse coding, comprises steps: 1) all images in the image data set to be processed densely extract block-level features respectively; 2) Randomly select several block-level features; 3 ) Establish a non-negative sparse coding model based on codebook block sparseness; 4) Use randomly selected block-level features to solve the codebook for the block-level features of the image data set; 5) Fix the codebook and perform all block-level features Non-negative sparse coding based on codebook block sparseness; 6) The encoding of each image in the image data set is integrated with the spatial pyramid maximum pooling method; 7) According to the objective function of the non-negative sparse coding model based on codebook block sparseness, find The codebook of the primary feature vector after the maximum pooling of the spatial pyramid of the image data set is obtained, and the primary feature vector after the maximum pooling of the spatial pyramid is subjected to non-negative sparse coding based on the sparseness of the codebook block to obtain the final value of each image Feature vector.

Description

Technical field: [0001] The invention relates to the technical field of computer vision image processing, in particular to an image feature extraction method based on codebook block sparse non-negative sparse coding. Background technique: [0002] Sparse coding has been widely used in various fields of computer vision (for example, image classification, image denoising, etc.), the traditional sparse coding method only requires that the coding is sparse (the so-called sparseness means that there are more zero elements in the coding), and There is no requirement for the structure of the codebook. Some recent experimental results show that: using a well-designed codebook can often achieve better coding results; non-negative coding often has better stability in applications. At the same time, some research results show that the image is composed of some local small pixel blocks, and all natural images can be combined by using a certain number of small local pixel blocks learned....

AI Technical Summary

Problems solved by technology

At the same time, some research results show that the image is composed of some local small pixel blocks, and all natural images can be combined by using a certain number of small local pixel blocks learned. The existing sparse coding method does not reflect this. kind of locality

The traditional method of extracting image features with sparse coding only encodes the block-level features once, and the image features obtained in this way are not highly discriminative in applications.

Images