Virtual Photograph Based Saliency Analysis of High Dynamic Range Images

Abstract

Computational visual attention systems detect regions of interest in images. These systems have a broad range of applications in areas such as computer vision, computational aesthetics, and nonphotorealistic rendering. However, almost all the systems to date are designed for low dynamic range (LDR) images and may not be suitable for analyzing saliency in high dynamic range (HDR) images. We propose a novel algorithm for saliency analysis of HDR images that is based on virtual photographs. Taking virtual photographs is the inverse process of generating HDR images from multiple LDR exposures, and the virtual photograph sequence has the capacity to more comprehensively reveal salient content in HDR images. We demonstrate that our method can produce more consistently reliable results than existing methods.