Salient Object Fast Detection In The Cluttered Scene

The goal of salient object detection is to find very different object in a scene. Itimitates human vision selective system which can filter out the backgroundinformation. In computer vision task, it provides rough position of target aspreprocessing model. It is always used in object tracking and recognition、imagesearch、image compression and so on.In reality, salient object is surround by cluttered background all most time. Butfew works focus on cluttered scene. This thesis just focus this problem and find out agood performance method to detect salient object in cluttered scene. Firstly, weanalyze many existing methods and work out a basic technical framework which iscombing frequency based method and contrast based method. We make someimprovement on both two methods in order to get high performance combinedmethod. In frequency based method research, we choose the newest method HFT andpromote its detection speed by using multi-scale thought. In contrast based methodresearch, we come up with a new local contrast operator based on center-surroundcontrast operator. Local contrast operator is used to detect small salient object andcenter-surround contrast operator is used to detect big salient object. It can providegood result with little detection time. Finally, we combine HFT and contrast basedmethod saliency map with mean add operator to get final saliency map. We providequantitative evaluation which shows the proposed method outperforms many otherstate-of-art methods.