Object Location Based On Saliency

In a variety of image analysis and processing, regions of interest is usually not all the contents of the picture, but only part of the image, therefore, a large number of researchers focus on human visual attention mechanisms and raised a number of model. For example, Laurent Itti et al presented model of visual attention. Computational model of visual attention based on anatomy, neuropsychology, cognitive science and other areas of research or theory. The mathematical model to simulate human visual perception system is focus of the digital image processing and computer vision, and other related direction of research.Computational model of visual attention can detect region of interest by finding a significant pixel of the image. In image retrieval, object detection, scene control, image resizing and other image processing tasks, computational model of visual attention detection in the region of interest focus on the calculation and analysis of such areas, compared to the image in all regions of the same priority given to treatment and reduce the computation process for the subsequent processing and provides great convenience.In this paper, the computational model of visual attention as a starting point, the discussion of the salient region extraction and object location technologies. This paper presents a method which integrates Saliency map and image segmentation. This method applied to the object location technology. The contributions includes:(1) this paper improve bottom-up saliency computational model. First, based on principles of human vision system, image is decomposed into brightness, color and direction of the three features. Each feature map make multi-resolution Gaussian pyramid. In the calculation of curvature calculated by adding the process to get the contour of objects, to obtain the corresponding image sequence. These image sequences are processed by Center-Surround operators in order to calculate the feature significant maps.Feature maps combine linearly into saliency map. Final saliency map is obtained by region-grow segmentation method. The improvement can make global structure of image clear. (2) Object location methods based on saliency map:the visual saliency map, based on the use of effective sub-window search algorithm has been significantly objects location. This object location method by Saliency map can not need for large training set.