Modified angular radial partitioning for edge image description in image retrieval

Abstract

An angular radial partitioning (ARP) method divides an edge image into M x N partitions in order to make its compact image descriptor. However, it inherently suffers from a boundary problem that arises when edge pixels exist near the partition boundaries. Proposed is a modified ARP (mARP) to reduce degradation of retrieval performance caused by the boundary problem. For comparison, image retrieval tests have been carried out using the above two methods. The results demonstrate that the retrieval performance of the mARP is almost twice as good as that of the original one.