Content-based image retrieval using direct binary search block truncation coding features

This paper presents a new image feature descriptor derived from the Direct Binary Search Block Truncation Coding (DBSBTC) data-stream without requiring the decoding process. Three image feature descriptors, namely Color Autocorrellogram Feature (CAF), Legendre Chromaticity Moment Feature (LCMF), and Local Halftoning Pattern Feature (LHPF), are simply constructed from the DBSBTC min quantizer, max quantizer, and its corresponding bitmap image, respectively. The similarity between two images can be measured using these descriptors under specific distance metric. The proposed method yields better image retrieval performance compared to the former Block Truncation Coding (BTC) and existing schemes under the natural and textural image database in the grayscale and color space. The DBSBTC performs well for image compression, at the same time, it gives an effective discriminative feature in the image retrieval task.