Angular Radial Partitioning for Edge Image Description

A novel approach for image representation based on geometric distribution of edge pixels is presented. Object segmentation is not needed, therefore, the in-put image may consist of several complex objects. The edge map is divided into M × N angular radial partitions and local features are extracted for these partitions. The approach is scale and rotation in-variant and also tolerates small translations and ero-sions. The extracted features are characterized by their compactness and fast extraction/matching time. They exhibit significant improvement in retrieval per-formance using ANMRR measure. Experimental re-sults show their supremacy using an image database initiated from a movie.