A patch-based local features and HOG integration model in context of visual categorization

The paper suggests a novel approach for visual key-points description. Our approach is basically integrate the descriptors of local features (e.g. SURF (Speeded Up Robust Features)) and HOG (Histograms of Oriented Gradients) using patch-based integration to enhance the distinctness between image objects. It was tested in the context of image categorization using bag of features model. The experimental results on Caltech101 dataset with 7 categories show that the proposed approach achieve the best performance over several state-of-art approaches. Also this paper investigate the impact of HOG patch-based integration on these methods in terms of accuracy and computational overhead.