Contour Based Multi-object Classification Technology

摘要

We propose a contour based feature descriptor for object classification. This method uses polygonal approximation algorithm to simplify contours and use adjacent lines to encode object contours. We demonstrate the high performance of the local contour descriptor within a powerful bag of fteatures classification scheme. Through extensive evaluation on PASCAL 2007 Visual Recognition Challenge dataset set, the test results show that this local contour descriptor has many advantages. It is simple and computation efficient. And it is easy to reuse in other frameworks.