The Saliency Detection Based On Region Contrast And Statistic Characteristics Of The Image

With the popularity of the digital products and the rapid development of Internet technology, the image has become an important source of information. How to extract interesting information quickly from an image by computer has become a focus on image processing and the field of computer vision in medical, milita and our life. Detection salient images inspired by biological mechanisms of Visual attention, the main task is to detect regions of interest or saliency regions in the image, and output a gray-scale image known as a saliency map. The bigger the saliency map pixel gray level value, the more likely belongs to the region of saliency, it can be assigned priority in the follow-up process computing resources to analyze the saliency region to reduce the processing burden on image analysis systemThis thesis Based on the biological Visual attention mechanisms analyzed the rationality and feasibility of saliency detection algorithm based on image region, Found that existing methods of saliency detecting can only detect general salient objects, but the recall and the precision can further enhance. To solve this issue, this paper propose an improved salient detect method based on regional features and statistical characteristics.First of all, This thesis using super pixel segmentation algorithm contrast of the image region. Then improving a block and chessboard distances replaced by linear combination of high computational complexity of Euclidean distance algorithm detecting saliency regional features of Euclidean distance algorithm makes it improved compared with the original detection method. And according to the spatial distribution of color variance to detect saliency map. Finally, the test results of the two models through the joint Gaussian model complement for saliency map. Let fusion method to detect salient targeting more accuracy and faster. Finally in the open test dataset MSRA-1000 using this method and a variety of significant methods to do a total comparison further validate the advantage of our method in accuracy of detection and time consumption.