Abstract: BOFs model is one of the most prominent, successful and used for the purpose of image classification. In spite of having many advantages such as very less complexity, ability to be scaled and generality, it scum to various drawbacks, which includes local descriptors provided limited semantic descriptors ,vulnerable structures depending upon single visual words and spatial weighting is inefficient. Numerous techniques have been proposed to nullify the effects of the above mentioned disadvantages, such as multiple descriptor extraction, (ROI) detection and spatial context modelling .Although these methods has contributed towards the improvement of BOFs model to little extent but still coherent integration scheme of all the modules is lacking to resolve the above mentioned problems, a unique framework with spatial pooling of various features is proposed in this paper. BOFs model is expanded on three aspects by the proposed model. First, SURF (speed up robust feature) descriptor is used which combines texture and edge based local and global feature together. Next, extraction of spatial context depending upon features required for midlevel image representation is done using geometric visual phrases. Finally, combination of effective and useful spatial weighting technique and smoothed edge map is used to capture the required features of image.
Keywords: Image classification, BoF Model, K-medoid clustering, Image matching, SURF.