Small salient target detection using overlapped sub window
Lei Ren, Chaojian Shi, Ran Xin · 2011
A small salient target detection method using overlapped sub window is proposed. Lab color space is employed in the method because the three components are unrelated. Then each of them is divided into overlapped sub window and frequency tuned saliency detection method is applied for each patch. The master saliency map is obtained by fusion of all local saliency maps. Within overlapped regions, the saliency is the result of competition among all saliency values. Compared with random sub window mean method, the advantages of proposed method are three folds. Firstly, no regions are ignored for saliency computation because the selection of sub windows is not random. Secondly, the proposed method avoids repeated computations for lots of overlapped unnecessary sub windows and the proposed method is of high efficiency. And thirdly, the competition strategy within overlapped regions substitutes direct addition in random sub window mean method, as a result, the saliency map demonstrates more accurate salient target region. Results of experiment show its better performance than frequency tuned method especially when there are lots of clutters in the sea background. In addition, results demonstrate that small salient targets can be detected and sea clutters are suppressed deeply.