Research On The Method Of Visual Word Generation Based On Fuzzy Concept Lattice

As a usual expressing way for images, bag of visuals is widely used in scene classification. But due to the “noisy” visuals, the accuracy of scene classification is significantly affected. As a useful tool for extracting uncertain knowledge, fuzzy concept lattice is benefit for getting the different hierarchical visual words by virtue of hierarchical characteristics and reducing the “noisy”visual words by virtue of multi-valued mapping characteristics. The paper focuses on generating method of visual words and scene classification by using fuzzy concept lattice. The main research work is as follows:(1) A method of producing visual words based on fuzzy concept lattice is presented. A context of multi-valued mapping for the BOV of training images is first produced. Then by using the hierarchical structure of fuzzy concept lattice,the “noisy” visual words are reduced through gradually reducing the membership function's value-interval, and an optimal value-interval is obtained.Finally the set of visual words exhibiting the scene category is acquired from the interval, and the effectiveness of this method is proved in the experiments.(2) A method of many value-intervals to produce the bag of visual words based on fuzzy concept lattice is proposed to avoid the deficiency of one optimal value-interval presented in the above method. First bag of visual words are built from different scenes. Then the best value interval of each category of training images is acquired after building different fuzzy concept lattices from different categories. Finally the visual words from each category are extracted in the different optimal value-intervals, and the classification accuracy is increased more efficiently. The experiments results prove its effectiveness.(3) On the basis of the above research work, a prototype system for scene classifying based on the fuzzy concept lattice is built by using MATLAB and VB.