Saliency-based stereoscopic image retargeting

Abstract

In this paper, we propose a new saliency based stereoscopic image retargeting method based on the characteristics of the Human Visual System (HVS). A new stereoscopic saliency detection method is designed by adopting low-level features of intensity, color, texture and depth. Besides, the viewing bias factors including center bias and depth bias existing in the HVS are adopted for stereoscopic visual attention modeling. Since the HVS is sensitive to edges in images, we fuse the saliency and edge maps to predict the visual significance of image pixels for image resizing. With the visual significance measure, we propose an image resizing method by minimizing the structure and depth distortion for stereoscopic image retargeting. Experimental results have shown that both our stereoscopic saliency detection and image retargeting methods can obtain better performance than the existing related methods on the public databases.