Image semantic retrieval using image fuzzification based on weighted relevance feedback

In this paper a new approach is presented for image retrieval using image fuzzification based on weighted relevance feedback on image feature vectors. The mind attributes of present work are the image retrieval using fuzzy approach and the relevance feedback by attention of the query image and the resemble images in the database. At the first the color features (Hue) and saturation (S) for each image by help of Fuzzy C means (FCM) algorithm are quantized to 20 bins. Then the feature vectors for each image are become fuzzy using KNN algorithm and for corresponding image, the 20 Gaussian functions have been regarded. The data scatter in around of each image from viewpoint of features by help of covariance matrix of each image is caused the retrieval of similar samples become excellent. The dominant point of this paper is a new approach for relevance feedback. In the relevance feedback by attention of weighting of image semantic groups by user, the image weight which belongs to each semantic group is changed. The obtained results on an image database show that our approach accuracy versa the number iteration with respect to [1] is better. The images using in this paper, are selected from Corel database and Simplicity project.