Shape feature descriptor using modified Zernike moments

Abstract

Shape feature extraction and description is one of the important research topics in content-based image retrieval. The paper presents a shape feature descriptor using the modified Zernike moments based on the Zernike moments with minimum geometric error and numerical integration error. This approach maps the region of interest of an image into a unit disk, and forms Zernike moments by computing projection of the mapped image onto Zernike polynomials. In addition, retrieval performance of the system is improved by introducing psychophysioiogical research results into computation of the Zernike moments. After having compared the traditional Zernike moments, the Zernike moments with minimum geometric error and numerical integration error, and the modified Zernike moments by experiments, it is found that the modified Zernike moments is better than the other approaches viewed from reconstruction. Furthermore, a system using the Zernike moments with minimum geometric error and numerical integration error has better retrieval performance than the systems using the traditional Zernike moments; a system using the modified Zernike moments has better retrieval performance than the system using the Zernike moments with minimum geometric error and numerical integration error from the view of retrieval. In addition, the retrieval performance of systems using the modified Zernike moment and the Zernike moments with minimum geometric error and numerical integration error can be improved when the given threshold increases slightly.