Concept of Doubly Supervision and ClassificationBased on Frequent Pattern Mining for ImageRecommender Systems

With the advent of databases that store hugeamounts of data, access to useful information has become as amajor challenge for users. Content-based image retrieval (CBIR)is one of the most important research areas in digital imageprocessing that automatically retrieves similar images to thetarget images and provides them for the users via the visualcontent mining and processing them. In this regard, many studieshave been conducted to enhance the accuracy of image retrievalsystems. But according to explosive growth of storage resourcesand lack of accountability systems for retrieving proper images,it is still considered as one of the most active research fields. Inthis paper, a method is proposed in order to cope with thechallenges mentioned that extracts the proper features of imagesby using the wavelet transform and combining it with the colorhistogram. Next, it searches similar images with the image queryand provides them to the users by using the concept of doublysupervision concept which uses the combination of clusteringmethods and classification based on frequent pattern mining. Theaim of this study is to enhance the accuracy in imagerecommender systems. To evaluate the proposed system,different classifications have been used to retrieve proper images.The test results suggest that the Multilayer Perceptron NeuralNetwork has the highest performance in terms of precision andrecall for our proposed method. The precision of the proposedmethod is 0.88 percent, this increased 0.13 percent than thecompared systems.

کلیدواژه ها:

Image recommender ، wavelet transform ، color histogram ، clustering ، frequent pattern mining ، classification