A novel fast scene classification method via DCT

摘要

Scene classification is a hot problem in the computer vision. In this paper, a novel rapid scene classification method is proposed based on the discrete cosine transform (DCT) domain. Firstly, we divided the whole image into several areas of the same size without repetition, do DCT transform with the size of the BB in each separate sub image area divided above. Secondly, scanning AC coefficients in each DCT block by three ways, modeling based on the correlation between AC coefficients in DCT block for extracting the feature vectors. Finally, with the feature vectors obtained previously, using one-vs-all Support Vector Machine train classifiers. Experimental results show that the proposed method is effective in image classification.