Object Recognition and Classification Based on Improved Bag of Features using SURF AND MSER Local Feature Extraction

Object recognition and classification is a challenging task in computer vision because of the large variation in shape, size and other attributes within the same object class. Also we need to consider other challenges such as the presence of noise and haze, occlusion, low illumination conditions, blur and the cluttered backgrounds. Due to these facts, object recognition and classification gained attention in recent years. Many researchers have proposed different methods to address the problem of recognition. This paper proposes a method for object recognition and classification based improved bag of features using SURF(Speeded Up Robust Features) and MSER(Maximally Stable External Regions) local feature extraction. Combination of SURF and MSER feature extraction algorithm can improve the recognition efficiency and the classification accuracy can be improved by spatial pyramid matching. SURF and MSER extracts the local features of an image and generate a image histogram codebook. Spatial pyramid matching is applied to this histogram, which improves the accuracy of classification. The experiment is conducted on Caltech 101 and Caltech 256 dataset.