Research On Algorithm Of Content-based Image Retrieval

The rapid development of network and computer technology has accelerated the pace of the new era of information,intelligence and technology.The number of data information is increasing rapidly,among which digital image accounts for a large proportion.It has become a serious problem to find the digital image information required by users in the huge database.Therefore,it is very necessary to develop a high-speed and effective technology for retrieving specific image information.In order to achieve this goal,the image must be represented by specific features.Color and texture are two important visual features of images.Therefore,the use of color and texture features for image retrieval is an effective image retrieval technology.In this paper,the method of block weighted histogram in HSV space is introduced firstly.The edge length of the image is segmented according to the 1:2:1 scale,and the color histogram of each sub-block is extracted in quantified HSV color space by non-equal intervals.Color features are represented by a cumulative weighted histogram.At the same time,the weight of different segments is improved,and the improved method is used as the color feature extraction method.Then the dual tree-complex wavelet transform is introduced,and the low frequency detail coefficient matrix of the three-layer dual tree-complex wavelet transform is extracted for the image.After the complex matrix is transformed into real matrix,the variance and entropy of each real matrix is extracted to be the texture feature information of the image at this scale.Finally,the color feature vector and texture feature vector are normalized by Gaussian model and fused to form a new feature vector.A dynamic weight matrix is set up according to the difference between different dimensions of features,and the similarity of different images is measured by weighted Euclidean distance.In order to avoid the mismatch or retrieval error caused by the heterogeneity among the feature extraction methods,this paper uses Gaussian model to normalize the distance,and finally uses the bubbling sorting method after fine tuning to sort out the distance.Experimental results indicate that the proposed method is more effective than the single feature retrieval method.