Image retrieval method based on multi-view local reconstruction preserving embedding

Abstract

The invention discloses an image retrieval method based on multi-view local reconstruction preserving embedding. The method comprises: extracting features of an image, to obtain feature matrixes of multiple viewing angles, using a k-nearest neighbor sample of samples in an original space to reconstruct the sample, through a reconstruction weight matrix to describe similarity among the samples andneighboring samples, mapping similar structure features in the original space to a low-dimensional subspace, constructing a similarity matrix according to the reconstruction weight matrix of each viewing angle, at that same time, constructing a set of non-negative weight coefficients to represent different weight contribution of the same sample from different viewing angles, introducing auxiliarycoefficients to fuse multi-view feature information, and through an iterative optimal solution method, obtaining common low-dimensional subspace embedding representation of multiple viewing angles, finally, performing similarity measurement sorting on a to-be-retrieved image and all images, to obtain a retrieved result. The method gives full consideration to consistency and complementarity information of multiple viewing angles, optimally solves embedding representation of low-dimensional subspace of an image, and improves performance of image retrieval.

Description

technical field [0001] The invention belongs to the field of multi-view learning, relates to image processing, in particular to an image retrieval method based on multi-view partial reconstruction and embedding preservation. Background technique [0002] With the increase in the amount of Internet image data and the richness of image information, more and more image noise information is generated. How to quickly and effectively mine valuable information in massive image data, quickly and accurately in massive images containing rich visual information Retrieving the images needed by users accurately is a research hotspot in the field of image processing that needs to be solved urgently. Since image data is often high-dimensional data in nonlinear space, and most current multi-view learning techniques cannot handle high-dimensional features of nonlinear space data well, the present invention seeks a multi-view Shared latent subspaces to efficiently solve problems in nonlinear...

AI Technical Summary

Problems solved by technology

Since image data is often high-dimensional data in nonlinear space, and most current multi-view learning techniques cannot handle high-dimensional features of nonlinear space data well, the present invention seeks a multi-view Shared Latent Subspaces to Efficiently Solve Problems in Nonlinear High-Dimensional Spaces

Images