Natural Image Classification Technology Research Based On Artificial Immune System And Content

Natural image classification has become a crucial technique in the fields of image intelligence, web image retrieval, video analysis, internet data filtration and so on, it has a promising application prospect. With the rapid development of the Internet, the number of the digital images increases explosively. How to classify natural images efficiently becomes a practical research. Currently, the research falls into two aspects: propose new classification theory and extract new information and features from images. The paper makes a research according to these two aspects:(1)A novel image classification algorithm based on artificial immune system (AIS) is presented. The algorithm adopts one-time presentation of antigens on the basis of Clonal Selection Algorithm (CSA) and resets the recognization rules. After all the antigens stimulating the antibody network, the antibodies of superior quality are selected to go through the clone and mutation. Considering the concentration of antibodies when selecting the antibodies, modifying adaptive clone and mutation rate and mutation expression, the algorithm is more capable of the searching optimal solution. The experiments on natural images show that the new algorithm can perform with faster searching speed and higher average accuracy.(2)This paper proposes a spatial feature of color based on information entropy. Firstly, extracting the global color histogram. Secondly, image was segmented unevenly and the color distribution was calculated, then the weighted blocked color distribution entropy (W-BCDE) was extracted in accordance with the features of human visual and symmetric properties of information entropy. Besides, the paper analyses how to determine the weight of each block and proves the low computational complexity of this method. The classification experiments using the artificial immune algorithm proposed by this paper show that the new features effectively overcoming the disadvantage of global color histogram not taking the color distribution into consideration. W-BCDE outperforms BCDE because it adopts weight to eliminate the negative effects brought by symmetric properties of information entropy.