Research On Multi-wavelet Image Retrieval

With the improving broadband networks and data storage technology, the number of digital image increase sharply. How quickly and efficiently find the desired image from the image database has become a key to the development of multimedia technology. Meanwhile, particularly there are a lot of compressed image, image retrieval in the compressed domain is a new research application area. This paper reviews the status of image retrieval research, in particular, in the transform domain and compressed domain image retrieval. On this basis, we conduct research around the multi-wavelet theory and multi-wavelets in image retrieval applications. In the multi-wavelet transform domain and compressed domain, the paper both proposed can improve the recognition rate of feature extraction methods. Specifically, the main contribution of this thesis in the following aspects:(1) According to the characteristics of multi-wavelet, we analysis the sub-band and the statistical characteristics between the different components by three different wavelet to image, and select multi-wavelets basis which is more suitable for the image retrieval method.(2) From improving recognition rate perspective, we overcome the sub-band only consider the limitations of the statistical features, and full use of multi-wavelet sub-band correlation. The paper proposes a new image retrieval methods based on the correlation between multi-wavelet sub-bands.(3) The paper introduces the multi-wavelet to set partitioning in hierarchical trees (SPIHT) algorithm, and analysis on the compressed stream of features of SPIHT coding based on multi-wavelet more. The method that extracts features from these compressed images achieves image retrieval only part of the decoder.The paper from the multi-wavelet transform domain and the compressed domain two aspects research image retrieval methods. In the transform domain, the paper proposes a new feature extraction algorithm based on multi-wavelet sub-band correlation. Tests show that combining sub-band approximation with detail sub-band image retrieval method is very effective. In the compressed domain, multi-wavelet instead of wavelet.The paper proposes a new feature extraction methods based on SPIHT compressed stream. For compression images, you can extract information directly from the lists in the not fully decode circumstances. This method has the advantage of reducing storage costs and computational complexity, especially for the Internet and dynamic database query and retrieval of images.