Modulation domain texture retrieval for CBIR in digital libraries

We present a new texture retrieval algorithm that, for the first time, performs content-based image retrieval (CBIR) in the modulation domain by computing powerful low-level texture features based on computed AM-FM image models. Performance of the new algorithm is analyzed with respect to competing methods where texture features are computed from Gabor filter magnitude responses. Our experimental results show that the new algorithm achieves a significant performance advantage. We describe how the new algorithm will be used for the texture component of a novel CBIR service called DIRECT, which is designed to provide image based searches in distributed digital libraries without the need for manual entry of annotations and image metadata.