Research On Target Detection And Contour Extraction Technology Based On Visual Saliency

Target detection and heterologous image contour extraction have a bread application prospect in such fields as target search, intelligent transportation and military. Target research and heterologous image contour extraction algorithm is researched in this paper on the basis of human visual attention mechanisms.First, in the view of the issue that classical Itti model cannot extract the salient target accurately and the model does not possess favorable commonality in different situations, the improvements are made to the model using the brightness contrast feature replacing brightness feature and proposing an adaptive weight adjustment method to combine the three features in order to obtain the optimum saliency map, which is more adaptive in salient target extraction of different situations. Comparing with classical Itti model, the salient target can be extract more accurate with the improved method.Second, in the view of the issue that usual method for car license detection is not adaptive in complex background, we proposed a new method based on the visual saliency and color information to locate car license plate. In complex background, we first extract objects with high saliency and then use the color information to locate the accurate license plate area. Comparing with other method, the accuracy rate of this method has improved with 13%, which shows the method we proposed is robust to interference and deformation of license plate area. Besides, the improved model is applied to detect the salient object in the sea, and the result shows that this method is also adaptive.Finally, in the view of the issue that usual threshold segmentation methods and usual cluster segmentation methods cannot extract the contour due to the interference of background information, a new segmentation method based on the saliency map is proposed. We segment images by clustering the saliency map and merging the small areas into large areas, and on the basis of segmentation, we can easily extract the contour of the object. The results of the experiments show that the method can extract the contours of visual, infrared images and SAR images that match the result of human visual attention.