Saliency detection based on weighted color contrast of image patch

Abstract

Image saliency analysis is an important research content in the field of computer vision. At present, the main method of saliency analysis is to measure the saliency of single pixel or regular image patch. It is easy to be affected by image texture, noise and other factors, and some important information is lost in the process of segmentation, which makes it difficult to extract salient objects from the image. Therefore, a saliency detection algorithm based on weighted color contrast of image patch is proposed. Firstly, the original image is divided into different size and non-overlapping image patch structure. Then, the color contrast of the image patch, the number of pixels included and the spatial distance between the two image patches are calculated. Considering the influence of spatial distance between image patches on saliency value, the weighted color contrast model of image patch is used to detect salient region. Finally, considering the influence of spatial distance between pixels on saliency value, the salient region is enhanced by calculating the distance between each pixel and the center of the salient region. In order to evaluate this algorithm, we use the largest publicly available data set in the world for testing. Experimental results show that the proposed method has better precision and recall rate, can significantly suppress the influence of complex texture and noise.