Multi-thread normalized non-negative sparse encoder based method for rapid feature representation of image

Abstract

The present invention represents a multi-thread normalized non-negative sparse encoder based method for rapid feature representation of an image. The method comprises the specific steps of: local feature extraction of an image: densely extracting SIFT features and color moment features of each image in a data set, randomly selecting a plurality of pairs of the extracted SIFT features and color moment features to obtain a codebook by using a K-means clustering method, and optimizing solution encoder parameters according to a relaxation model based on the multi-thread normalized non-negative sparse encoder by using the codebook obtained after solution; and a test phase: using an encoder obtained in a training process to only extract the SIFT features of one input image during feature representation of the image, using the encoder to calculate coded coefficients of the SIFT features, and integrating all the coded coefficients according to a space pyramid maximized pooling manner, wherein an obtained high-dimensional feature vector is a feature vector of the image. The obtained image feature representation is further used in intelligent analysis application of classification / retrieval and the like of various images.

Description

Technical field: [0001] The invention relates to the technical field of computer vision image processing, in particular to a fast image feature representation method based on a multi-cue normalized non-negative sparse encoder. Background technique: [0002] Biological studies have shown that the response of mammalian primary visual cortex to external stimuli is sparse, that is, only a few neurons are activated, and the corresponding coding should be sparse coding. Sparse coding, in layman's terms, represents a signal as a combination of a set of bases, and requires only a few bases to reconstruct the signal. Sparse coding has been widely used in computer vision, image signal processing and other fields, such as signal reconstruction, signal denoising, image feature extraction, and classification applications. [0003] The traditional sparse coding method is based on reconstruction in the sense of minimum mean square error, that is, to make the reconstruction error as small ...

AI Technical Summary

Problems solved by technology

The exact solution of normalized non-negative sparse coding has high complexity and is not suitable for large-scale image applications

Images