An novel algorithm of progressive texture feature extraction in the wavelet domain from the compressed bit-streams of images

The massive data of images are usually stored and transmitted in compression format, which makes the processing of the images in compressed domain significant. Unfortunately, the high computational cost makes the current feature extraction methods of the compressed domain inefficient. This paper proposes a novel texture feature extraction algorithm named WETFC (Wavelet Energy Texture Feature Based on Compressed Bit-stream) from the compressed bit-streams of remote sensing images. Firstly, the compressed bit codes are classified into three types, namely Numeric Code (NC), Sign Code (SC) and Identification Code (IC). Secondly, the wavelet sub-bands of the bit codes are identified. Thirdly, the energy value that every NC code contributes to is calculated. We do not need to decode the compressed bit-streams and the progressive characteristic of the compressed bit-streams is kept as well. The experimental results demonstrate that the WETFC algorithm is more efficient and performable than the general method.