The Research And Application Of Image Retrieval Based On Multi-Feature

With the rapid development of internet and multimedia technology, rapid and accurateretrieval in the massive image has become the research focus. With its efficiency andreliable characteristics, content-based image retrieval is widespread concerned and grow fast,However, as image information is very rich, single feature is not sufficient to express animage. Therefore, image retrieval based on multi-features becomes the important research incontent-base image retrieval.Based on the related research at home and abroad, this article discussed several keyproblems in the image retrieval based on multi-feature. The main research work includes:1.On the basis of multi-features fusion method, a new image retrieval method thatintegrates BTC color moment and GLCM is proposed. First, we proposed a color featureextraction algorithm based on BTC color moment. On the basis of BTC, we improve thetraditional color moment. In our method, each image is divided into several image blocks.Then the pixels of each block are divided into two sets according the threshold. Last,compute the color average in two sets. In order to overcome the shortcoming of singlefeature retrieval, we integrate color feature and texture feature in the process of imageretrieval, the color feature is demonstrated by BTC color moment and the texture feature isextracted by GLCM. The results of experiments show that the retrieval results obtained fromcombined features are better than retrieval obtained from single feature.2.A new image retrieval algorithm based on gradient texton coherence vector isproposed for fusion of multi-features. The algorithm compute the edge gradient in theModified HSV color space first, and then gain gradient texton map by scanning the gradientimage using the special texton types. Next, the texton pixels are combined into thecoherence set, the other pixels are the non-coherence set. At last, the feature vector of theimage retrieval is represented by color auto-correlogram in two sets. Experimental resultsshow that the proposed algorithm can combine color, texture, shape and spatialcharacteristic effectively, and have valid precision and recall.