Region-Based Image Classification

Proposes a region-based feature descriptor for general object classification. The method uses polygonal approximation algorithm to simplify region boundaries, and uses a simplified SIFT descriptor to describe region appearance features. Demonstrates the high performance of this region descriptor within a powerful bag of words classify scheme. Through extensive evaluation on PASCAL 2007 visual recognition challenge dataset set. Test results showed that this region descriptor has many advantages. It can capture both shape and appearance features. It is simple and computation efficient, and is easy to reuse in under other frameworks.