Visual Object Categorization Based on Orientation Descriptor

The demand of new fast technology and image investigation in many applications has made managing visual object categorization techniques extremely important. The main problem of visual object categorization is the semantic gap (categorization problem). Currently, several researches show that using a texture feature could improve the categorization problem especially when using orientation descriptors. Mainly, in this research the edge histogram descriptor has been selected to extract the texture feature. Obviously, the main demerit of using this kind of texture descriptor is it uses single orientation to extract the texture feature. Therefore, the Gabor filter has been proposed to improve the performance of this descriptor by constructing different feature maps based on different scale and orientation. To demonstrate the performance of the proposed method, the first 20 classes of the Caltech 101 dataset have been used. Moreover, we compared the performance recognition of the proposed method in two different domains, namely spatial and frequency domains. Finally, the result shows that the proposed method in the spatial domain outperforms the proposed method in the frequency domain. This is because of losing some of the basic raw data though using Fast Fourier Transform algorithm in converting the system to the frequency domain.