Image Saliency Detection Algorithm Based on Spatial and Frequency Domain

Detection of visually salient image regions is important for object segmentation, adaptive compression and object recognition. Because of the complex colour, complex surface texture and changeable background in the image, the quality of traditional image saliency detection methods is not high enough to exactly reflect the saliency information of the image. The test results of traditional methods cannot clearly point out the outline of the salient object. Meanwhile, neither do they highlight the most prominent object, nor do they uniformly highlight the entire object. These shortcomings make the saliency map unreliable and inefficient in various applications. To solve these problems, we introduce a method for salient region detection that outputs high resolution saliency maps with high brightness salient objects. Firstly, the image is filtered based on contrast sensitivity function to suppress the redundant image information. Then the saliency map is extracted in both spatial and frequency domains. Finally, the two saliency maps are weighted together to obtain the final saliency map.