Digital Signal Processing Reference
In-Depth Information
Wavelet Domain Distributed Information Entropy
and Genetic Clustering Algorithm for Image Retrieval
Kamil Moydin
1
and Askar Hamdulla
2
1
Institute of Information Science and Engineering Xinjiang University, Urumqi 830046
2
College of Software, Xinjiang University, Urumqi 830008
askar@xju.edu.cn
Abstract.
After segmenting the image into several sub-images, each sub-image
is taken through three level wavelet transform, and then the texture images are
obtained. Meanwhile, the distributions of each sub-image's information entropy
are calculated. Such a way, both the global wavelet texture information and the
spatial distribution of information entropy are effectively used as the main re-
trieval characteristics. On this basis, the genetic clustering algorithm used for
the image clustering, and the likelihood between the query example image and
corresponding image's cluster center is calculated. Experimental results show
that the method presented in this paper has good retrieval performance.
Keywords:
Image retrieval, wavelet transform, wavelet histogram, wavelet
information entropy distribution, genetic clustering.
1
Introduction
In recent years, because of the urgent needs, CBIR (Content-based Image Retrieval)
technology has been get widespread concern and rapid development. Because of good
spatial frequency characteristics and transform mechanism consistent with human
vision system, the wavelet transform occupied an important position in the new gen-
eration of still image compression standard (JPEG2000) and moving image compres-
sion standard (MPEG24). So, study image indexing technology based on wavelet
domain has great significance. This paper proposes an effective image retrieval new
method based on wavelet information distribution entropy, and its main idea is: on the
basis of image segmentation, calculate the wavelet information entropy of each sub
image by applying wavelet and information entropy theory, and take it as image fea-
tures to do image retrieval. This method is characterized by: (1) Feature extraction
from the segmented sub image greatly reduces the computational complexity. At the
same time, the representation of image features has good tightness. (2) It could effec-
tively reduce the feature dimensions of images, and then the speed of retrieval is
greatly improved.
