Using Wavelet Texture And Shape Features Of Image Retrieval And Systems To Achieve

Content-Based Image Retrieval (CBIR) technology as a new technology has been applied and developed broadly. In CBIR technology, firstly, the content of an image is represented by its low-level visual features, then, these features are expressed into characteristic vectors to accomplish the retrieval. Apparently, it is very important to extract and express these low-level visual features accurately and effectively.This paper makes a study of how to extract the texture and shape features accurately and effectively with the image processing, computer vision and database technology. Improved algorithms of the texture and shape features are proposed utilizing the Multi-Resolution Analysis of the wavelet.In the extracting of texture feature, in order to avoid the disadvantage of the usual wavelet which doesn't have the property of rotation, shift and scale invariance, a new method which applies the first order angular moments to complete the rotation correction is proposed in this paper, then, the wavelet which has the property of shift and scale invariance is applied to decompose the orientation-normalized image. The invariant wavelet decomposition can be got. The mean and standard deviation of the wavelet coefficients are applied to describe the texture feature.In the extracting of shape feature, firstly, the Multi-Resolution edge is extracted utilizing the wavelet absolute maximum value. After that, a vector group which is constituted by the seven Hu invariant moments and the five promoted moments is applied to describe the shape feature.Only one aspect of the image's feature can be reflected by the single feature-based image retrieval. The advantages of these features can be well combined to complement one another by combining several features. So, on the basis of the improved texture and shape features' extraction and retrieval, an algorithm which combining the texture and shape features is realized in this paper.Through studying and analyzing image retrieval system model, a prototype image retrieval system with Visual C++6.0 is implemented in this paper. The improved texture, shape and combined-feature algorithms are tested using the system. The experimental results show that:a higher precision and stronger robust can be got by using the improved texture algorithm and shape algorithm; the result using the multi-feature image retrieval is much better than the result using the single feature-based.