JPEG2000 Image Retrieval In Compressed-domain

Visual information is being extensively used in various applications. Although it contains a wealth of information, the corresponding to the amount of data is very large, hence, there are large image databases and in the real time applications, it is a challenging issue mainly in image retrieval.Although the effect of the existing content-based retrieval (CBIR) is relative good, it needs to be uncompressed the image fully, and then extracts the characteristics used to retrieve, which consumes large amounts of memory space and could not be used on the network in real time, so many people proposed the technology of compressed domain-based image retrieval. With many compression standards are produced, they began to show their importance increasingly.Because of the high compression ratio of JPEG2000, the clarity of uncompressed image, progressive transmission and restoration of the region of interest (ROI) of high-quality, JPEG2000 will replace the existing JPEG compression standard, became a new generation of compression standard. So the technology of image retrieval based on JPEG2000 compressed domain has very good practical value.This thesis analyzes the development of image retrieval technology in compressed-domain and the JPEG2000 static image standard, and then proposed two methods of JPEG image retrieval in compressed-domain.①First, an image retrieval technique in JPEG2000 compressed domain using wavelet coefficients is proposed. We use Context Labels rationally. We judge the effectiveness of each code block (0, 1) in every sub-band by the values of the coefficient, and use nine Context Labels to compute the effectiveness of code blocks in each sub-band, and the probability of the nine Context Labels in each sub-band are used to calculate the similarity. Experimental results show that this method has a good effect, especial to image which have texture characteristics significantly②Second, an image retrieval technique which used the packet header information is proposed, According to the special organization of code stream, we can extract the feature including the number of non-zero bit-planes, the number of coding passes and the code block length from the packet header . we use the non-zero bit-planes information and the number of coding passes as two indexes, and the length of sub-band for the third index, this paper proposed the concept of sub-band compression ratio: the length of sub-band streaming to number of sub-band coefficients, as a third index. Experimental results prove the retrieval have fast speed, that because this method does not need to consume time on the JPEG2000 entropy coding and inverse quantization decoding stage which are the most complex.