Research On Content-based Image Retrieval Method In Compressed Domain

With the development of image compression technology, more and more image using a compressed format for transmission and storage. With the popularity of smart phones, digital cameras and other digital devices, a variety of image compression standards such as: JPEG,MPEG has been widely used in daily life. Content-based image retrieval became one of the hot spot research for nearly 20 years.Traditional content-based image retrieval algorithms dealing image in the pixel domain, which needs to decode the image file from compressed domain to pixel domain and extracted the image feature in the pixel domain. There are two problem in feature extraction in the pixel domain : on the one hand, the information has been lost in the lossy compression cannot be restored even decode the image to pixel domain; on the other hand, image decompression processing is time consuming, and there’re more data to deal with in the pixel domain.The paper focus on feature extraction for JPEG image in compressed domain, combine the image feature extraction with image retrieval technology, make some study on the area of image retrival in compressed domain:1. This paper propose a statistical texture feature extraction method for JPEG image in the compressed domain, on the basis of previous research. We analysis the DCT coefficient in the three channel in YCb Cr color space separately, DCT blocks distribution histogram feature is computed in luminance channel, and the AC coefficients distribution histogram of the DCT block is computed in two chrominance component.2. We applied this feature extraction algorithm to image retrieval in the Corel1000 image database, and compare with two more existing compresseddomain image feature extraction algorithm. Experiments show that the proposed statistical texture feature extraction of JPEG images based on DCT coefficient algorithm has achieved good results, with strong practical applications.