Image Retrieval Using Color And Texture Fused Features

Image contains abundant information,and plays an increasingly important role in Human life and work.Moreover,the unceasing innovation and wide spread application of the Internet and multimedia technology lead to sharp increment of the number of digital image.Image retrieval has gradually become an important way for people to analyze and utilize image resources,and has been widely used in many fields.However,as the fundamental and core technology of image retrieval,research of the low-level feature extraction has not yet reached a mature state,and is still a popular and difficult research issue.This thesis focuses on the low-level feature extraction in content-based image retrieval,especially the extraction of color and texture feature.The main research work is as follows:(1)Study and propose the feature extraction algorithm of color fuzzy correlogram(CFC).The existing algorithm of color auto-correlogram(CAC)only consideres the distribution of the same color in image,ignoring the relationship between different colors.And the limited extracted color information leads to the poor image retrieval effect of the CAC.Therefore,the thesis proposes the algorithm of CFC.The similarity value between colors is added in the CFC based on the CAC,and the value is calculated using the ambiguity function defined in this thesis.Contrast experiments show that the image retrieval time of the CFC is slightly higher than the CAC.But the retrieval accuracy of the CFC on the Corel-1K image library reaches 69.60% and is about 7% higher than that of the CAC.(2)Study and propose the feature extraction algorithm of color layer based texture element histogram(CLBTEH).On the basis of the existing texture feature extraction methods,especially the structural texture analysis methods,considering that it is difficult to extract texture elements and their arrangement rules directly from general images,this thesis identifies 16 texture elements of size 2×2 based on binary images to form a complete set.These texture elements are taken as basic units of binary images and the texture element histogram is used to describe the texture feature of binary images.And then,the algorithm of CLBTEH is proposed to extract the texture features of general images.Before feature extraction,images need to be binarized through color quantization and image stratification.And in this way,the color and texture information can be fused together during the feature extraction process to improve the quatility of image retrieval futher.The simulation experiments based on Corel-1K image database show that,within short retrieval time,the highest accuracy rate of the CLBTEH reaches 76.92%,being 1.5%-6.5% higher than that of the other several comparative algorithms.(3)Study the image retrieval using color and texture fused feature.The CFC algorithm extracts the color feature and the CLBTEH algorithm is mainly used to extract the texture feature.In order to achieve higher image retrieval accuracy rate,these two features are combined together.The experimental results show that the retrieval accuracy rate after the combination is up to 78.71%,and 1.8% higher than that of only the CLBTEH,9.1% higher than that of only the CFC,about 3.7%-7.6% higher than that of the other several comparative algorithms.