Geometric invariant object segmentation method based on sparse independent shape component representation

Abstract

The invention discloses a geometric invariant object segmentation method based on sparse independent shape component representation. The method comprises the following steps of: firstly, carrying out log polar coordinate transformation on a training set; carrying out independent shape component extraction on a transformation result; on the basis of an independent shape component, constructing the first mapping convex shape set of an independent component space; on the basis of the extracted independent shape component, establishing a second mapping convex spare shape subset; on the basis of the first mapping convex shape set and the second mapping convex spare shape subset, constructing a sparse independent shape component representation model; combining with the above sparse representation model to construct an object segmentation model with a level set energy constraint item; adopting a gradient descent method and a soft threshold value method to solve a target function; and utilizing an evolution curve to realize image object segmentation. By use of the method, the problems of sensitive geometric deformation, poor shape representation ability and low segmentation accuracy of an existing object segmentation method based on sparse representation are overcome, and the problem that the sparse representation model has poor object segmentation ability under a situation of poor shape alignment or unknown deformation parameters can be solved.

Description

technical field [0001] The invention relates to the field of image object segmentation and sparse representation, in particular to a geometrically invariant image object segmentation method based on sparse independent shape component representation. Background technique [0002] Object segmentation is one of the core links of image recognition, image classification and image understanding technologies. The existing models include segmentation methods based on energy functionals, segmentation methods based on graph theory, and Markov random fields. In image target segmentation, if the target shape is similar to some training samples in the training shape set, a model can be constructed to search for similar samples in the training set, and the sparse linear combination of similar samples can be used to approximate the target, which is beneficial to supervised target segmentation. Therefore, object segmentation technology based on sparse shape representation has recently beco...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved by the present invention is to provide a geometrically invariant image target segmentation method based on sparse independent shape component representation. Poor, the problem of poor segmentation effect

Images