Image Saliency Detection Based On Geodesic-like Features And Boundary Contrast Map And Its Application

Image saliency detection is very important for the subsequent processing.It is a current research hotspot to locate the saliency region accurately and efficiently from the vast amounts of images.For the images with a simple background and a single object,most of the algorithms can obtain satisfactory detect results.However,for images with a complex background and low contrast between background and foreground,still can't effectively extract the objects.This paper proposes an image saliency detection method based geodesic-like features and boundary contrast map.The main work and innovations are as follows:This paper proposed an image saliency detection method based on geodesic-like features and boundary contrast map.To the existing problems in the global and the local contrast methods,this paper proposed an image boundary contrast algorithm.This method compared background regions with other regions in the image and obtained boundary contrast map,then fused geodesic map with color contrast,and used the geodesic map without color contrast as the weight to calculate image saliency map.The ROC curve with distance(DROC curve)is proposed to evaluate the efficiency of image saliency methods.Classical ROC curve compares the area difference between ground truth and object in saliency map which didn't reflect the human perception.This paper added distance to ROC and proposed a distance ROC curve,which makes the evaluation result closer to the human perception system.This paper compares the proposed method with eight state-in-art saliency detection algorithms in three public image databases.Including DROC curve four evaluation methods are used as evaluation metrics.The experiments show that the proposed method has better performance which can extract image saliency regions more accuracy and has obvious advantages in images with low contrast between background and foreground.The proposed method is applied to image resizing and image compression.In seam carving technology,the saliency map is used as importance map to resize the image to different sizes.For JPEG2000 image compression,the saliency region is used as ROI;then image is compressed.SSIM and MSE are used to as image quality assessment indexes.The experiments show that the proposed image saliency method has better performance than the other eight state-in-art methods in the two applications.