Image feature fusion and clustering collaborative expression method and system for intrinsic manifold structure

Abstract

The invention discloses an image feature fusion and clustering collaborative expression method and system for an intrinsic manifold structure, and the method comprises the steps: inputting a group ofsimilar matrixes of an image in different feature spaces, and extracting a sample manifold structure from the group of similar matrixes through employing a nonlinear fusion operator based on a tensorproduct; further quantifying the influence of the noise intensity in the input similar matrix on the manifold structure through weight learning, and enhancing the robustness of the manifold structureobtained through fusion on noise; applying neighbor constraints and Laplace low-rank constraints comprehensively, thus the locality of a manifold structure and the consistency of cluster distributionare guaranteed, and collaborative expression of feature fusion and clustering is achieved. According to the image feature fusion and clustering collaborative expression method of the intrinsic manifold structure, detail information of the image data corresponding to the topological structure in different feature spaces can be acquired, and the method has high universality and robustness, and has the advantages of noise interference resistance, high clustering precision and the like.

Description

technical field [0001] The invention relates to image feature fusion and clustering technology, in particular to an image feature fusion and clustering cooperative expression method and system of a constrained essential manifold structure. Background technique [0002] In the current era, images have become a major information carrier, playing an increasingly important role in social development and economic construction. Using statistical or machine learning methods to mine different types of features in images makes it possible to observe image characteristics from multiple angles and distinguish samples with different labels in images with high precision. Compared with a single feature, multiple different types of features provide more angles and different levels of sample information. Through mutual support, supplementation, and correction, they can provide essential correlation information between image data. It has been used in hyperspectral remote sensing images. Ana...

AI Technical Summary

Problems solved by technology

At present, this type of method mainly has the following problems: First, the operator uses the following assumptions to mine the relationship between samples, that is, if the similarity between samples a and b, c and d is higher, then samples a and c, b The stronger the correlation between and d

However, this assumption is not always satisfied in practical applications, leading to the diffusion of some wrongly associated information; second, the information diffusion of this operator in different feature spaces is global, which is different from the data in practical applications. is locally correlated with a contradictory condition

Images