A Global Image Feature Construction Method Based on Local Jet Structure

Abstract

This article presents a novel and robust feature descriptor called the multi-scale autoconvolution on local jet structure (MSALJS), which is quasi-invariant to affine transformation. The MSALJS, a global image feature descriptor, is based on the derivatives that describe the image local structure to compute the multi-scale autoconvolution moment. Experimental data demonstrate that the MSALJS can be used in practical applications in which the object is deformed in various ways, such as particular occlusion, view angle change, and so on.