Incorporating shape into histograms for CBIR

摘要

This paper describes an indexing system for use in Content Based Image Retrieval. The standard colour histogram approach is simple, efficient, and robust. However, it does not include shape information, which leads to problems (e.g., many-to-many mappings). To remedy this, we use additional features in an attempt to incorporate shape and textual information to the index key. Our experiments showed that the combination of colour, texture, distance and orientation histograms gave approximately 10% improvement of recall over the standard colour histogram.