Content Based Image Retrieval Based On Color And Texture Feature

Along with the rapid development of multimedia and network technology, thereare more and more digital images on the internet. In order to obtain the desiredinformation quickly and efficiently from a large number of images, people come upwith Content-based Image Retrieval(CBIR) which has become a hotspot issue in thefield of image processing. One of the core technologies of Content-based ImageRetrieval is the Low-level Features based image retrieval. color and texture features aretwo important and widely used low-level feature of image.This paper focuses on the extraction of color and texture features for imageretrieval, especially the extraction of color feature, texture feature and combining themfor image retrieval. The research content is as follows:①In order to overcome the defects of using probability density function to modelthe wavelet subband detail coefficients. This paper proposed a new texture imageretrieval method by extending the Refined Histogram(RH) to model the magnitude ofcomplex coefficients of double density dual tree wavelet transform(DD-DT CWT). Theadvantage of this method is that it is a combination of the efficient, effective RH and theshift-invariant, anti aliasing, multi-directional DD-DT CWT. The experiment resultsshow that the proposed method yields higher retrieval rate than using the GeneralGaussian Density(GGD)model to fit with the real part or imaginary part of coefficients,and than using the Gamma PDF fit with the magnitude of complex coefficients.Experimental results on Brodatz texture dataset showed that the accuracy is increasedabout2%-9%.②It is hard to attain satisfactory retrieval results by using only color or texturefeature for image retrieval. Because in general, an image contains various visualcharacteristics. This paper proposed a new color image retrieval method by combiningcolor and texture feature. In which the color feature is presented by dominant colordescriptor extracted by linear block algorithm and texture feature is presented byRefined Histogram(RH). Dominant color descriptor provides an effective, compact, andintuitive representation of colors presented in an image and linear block algorithm isefficient in color quantization and computation. Experimental results on Wang databaseshowed that the average accuracy is increased about5.49%than using only color asfeature for image retrieval. and about4.43%than using only texture as feature for image retrieval.