Scale Invariants of Radial Tchebichef Moments for Shape-Based Image Retrieval

摘要

Region-based descriptors often use moments to describe shapes. Recently, the discreet radial Tchebichef moment descriptors have been proposed. The radial Tchebichef moments are invariant with respect to image rotation. In order to achieve the scale invariance, researchers resort to resizing the original shape to predetermined size. This traditional scheme of scaling is time expensive and leads to the loss of some characteristics of a shape. Therefore, moments derived using the traditional normalization scheme may differ from the true moments of the original shape. In this paper, a simple yet powerful scheme has been proposed to derive a new set of scale invariants of radial Tchebichef moments. This scheme uses the area and the maximum radial distance of a shape to normalize the radial Tchebichef moments. The MPEF-7 scale-invariant database is used to evaluate the performance of the proposed scheme against four commonly used shape descriptors.