Content Based Image Retrieval Based On The Color Of The Second Fuzzy Clustering

At present,the content-based image retrieval technology is a very active research field,in which one of the most important contents is the image color feature extraction and similarity calculation.In the past,the color characteristics of image processing algorithms focused on the use of computers for statistical analysis and classified information,but relatively neglected the visual characteristics of human observation of images.Meanwhile,redundancy of color quantization existed in the algorithm outcome; it was difficult to deal with boundaries of color quantization,and color quantization was also incompatible with the human visual characteristics.In this paper,based on the original algorithms and from the subjective human perception,a new algorithm is brought up with the combination of fuzzy theory in order to extract the image which can reflect people's perceptual characteristics better,so that the human similarity calculation can be in accordance with the habit of human image observation.First of all,this article compares the color quantization algorithm in common use and puts forward a new color quantization algorithm based on visual features by applying the previous way of calculating.In the HSV color space,the proposed method is more reasonable in the visual aspect.The quantification boundary is made fuzzy with the combination of fuzzy theory to conduct the first color quantization,and then the result of color quantization is given a second cluster.Experiments show that such a quantitative approach is in line with the human visual features and at the same time can effectively reduce the redundant information in the color histogram.Secondly,this paper shows the improvement of the traditional color similarity comparison algorithm based on the color quantization algorithm,a new image based on the main color of retrieval algorithms.The improved concept is to add weight value to colors according to the proportion of the color in images during the calculation of similarity,so that the colors with high proportion can make more contribution to similar results.Then,this paper presents the dynamic clustering method based on fuzzy equivalence relation,and proposes the classification algorithm based on the main color images on the basis of color quantization in the first part.Color image is quantified according to 23-color,the color characteristics of an image can be represented by one-dimensional array consisting of 23 constituents,and the sample set can be expressed as a matrix of 23.In this paper,a matrix consisting of sample set is clustered in accordance with the dynamic clustering method of the equivalence fuzzy relation, and introduces the method into the application of image classification.Finally,the establishment of image database is provided by Corel Photo CD and the Department of Computer Science and Engineering University of Washington image database.The performance of the algorithm is verified considering C# and SQL Server as a developed environment,and the new algorithm is compared with the original one.The experimental results show that image retrieval and accurate recall rate of the proposed method in this article are quite satisfactory and image classification can achieve the expected results.