Clustering method based on group sparse optimization

Abstract

The invention provides a clustering method based on group sparse optimization, and the method comprises the steps: firstly carrying out the processing of data, and obtaining a similarity matrix targetmatrix, an error minimum item and a sparse constraint item between data set samples; secondly, constructing an optimization model based on group sparse constraints, wherein the purpose of the optimization model is to suppress the noise influence by using more powerful group sparse constraints; thirdly, providing an optimization algorithm based on an Alternating Directional Multipliers (Alternating Directional Multipliers), and employing the optimization algorithm for quickly solving the constructed optimization model; and finally, a rapid optimization clustering algorithm is provided, which aims to merge redundant clustering results and further improve the performance. According to the method, each sample is constrained to be approximately represented by only one sample, so that the algorithm robustness can be effectively improved; on the other hand, spectral clustering analysis does not need to be carried out on the obtained target matrix, so that an end-to-end clustering effect is achieved.

Description

technical field [0001] The invention relates to the technical field of digital image forensics, in particular to a clustering method based on group sparse optimization. Background technique [0002] Digital image equipment forensics technology refers to the technology of judging its imaging equipment only by the image content, which has important research value in the fields of image processing and information forensics. The early methods are designed to judge whether the image to be detected is shot by a specific camera, and are widely used in copyright protection, evidence detection and other scenarios. However, since such methods need to obtain the identification features of the imaging device in advance, such methods have great limitations in practical applications. At present, given a set of images, it is more feasible to determine which images were captured by the same camera and which were not. This scheme does not depend on any prior knowledge, and can be applied t...

AI Technical Summary

Problems solved by technology

[0007] However, there is still a certain distance between the above-mentioned prior art and practical application.

For example, in the method described in Document 1, some features with severe noise interference will cause cumulative errors, thereby reducing the reliability of the traceability results

In addition, in the method described in Document 2, when faced with a large-scale data set, the method needs to integrate a large number of sub-clustering models, which seriously reduces the recognition efficiency

Images