Salient Region Detection Based On Low-level Features And High-prior

With the rapid development of information technology and the popularity of mobile communications equipment, mankind has entered into the "map reading age", image and video data are exploding. Because the image equipped with briefness, vividness and convenience, which has been applied to many fields, it is bound to increase computer burden. How to deal with the image data effectively has become the focus of people’s research. According to the principle of human visual attention mechanism, a lot of image saliency region detection algorithms are constructed by using the mathematical model to process priority the salient regions of the image, which would bring convenience for the image compression, target search or other follow-up work. In order to further improve the accuracy, this paper carried out a study on the significant area detection algorithms.Firstly, use two kinds of segmentation algorithms to obtain the image primitives. Then, analyze the effects of the low-level features to the salient regions, put forward some new definitions of primitive color contrast, primitive brightness contrast and color space distribution characteristics.Secondly, taking the limitations caused by simple use of low-level features into account, some high-level prior knowledge are explored, such as boundary generally is the background information, and background information related to border connectivity strength is greater than the target object communicates with border strength; center prior stands for the salient regions are close to the image center than the non-salient regions; warm advantages, goals characteristic and other senior prior knowledge.Thirdly, the combination of the significant features can improve the detection effect. Therefore, a variety of feature fusion strategies are proposed to generate high quality images, as well as multi-scale space is used to enhance the detection results of the image.Finally, evaluate the methods proposed by the paper in three international public data sets: ASD, BSD and SED2. And elaborate the advantages and disadvantages from quantitative and qualitative two aspects.