Research Of Image Retrieval Based On Brushlet Transform

With the multimedia and computer digital technologies high-speed developed, Now, as an important part of multimedia information,digital image has become a form of media which used widely in science and technology, education,business and so on.Extract the required images for user is an important aspect of information retrieval,which has broad application prospects.Therefore,image retrieval has important value of practic and research.Currently, content-based image retrieval is a hot research,it mainly uses basic visual features of the image itself exist objectively to retrieve.Feature extraction is the most crucial part in content-based image retrieval.Texture has rotation invariance and strong resistance for noise.So it is an important feature for image retrieval.Multiscale geometric analysis originated from and higher than wavelet.It is a new image analysis tool,which is multi-directional and anisotropic.It could extract image feature more effectively. Brushlets is one effective image analysis tool to solve the angle-resolved problem.Therefore,this paper studied from Brushlets of Multiscale geometric analysis to describe texture information in images,and carried out two parts study on image retrieval.1.This paper proposed a image retrieval algorithms based on complex fea-ture from Brushlet transform.Since Brushlets is a complex function and is multi-directional,then we extracted energy and phase characteristics to describe the tex-ture information.Experiments on Brodatz texture image database show that this retrieval results are better than Gabor transform and wavelet transform,the algo-rithm effectiveness on the texture image retrieval was verified.At the same time, the impact of Brushlet transform decomposition levels in the retrieval results was analyzed.Further,the retrieval results and reasons for different kinds of texture images was discussed. 2.This paper proposed a new image retrieval algorithm based on hybrid feature of color and texture in Brushlet domain.In order to avoid the limitations for losting color information of color image,we combined color feature after extracted image texture feature,then quantified color space on HSV space which is more in line with human's vision,and described color feature with color histogram.The over-all characteristics statistics of the image itself are obtained and classed by k-means clustering.Further,a new weight operator which can weight for each image adap-tively was seted as the differences between the images.Experiments shows that this method retrieved similar images more priority than single features and traditional composite character,it improved color image retrieval performance significantly.