Perceptual based stereoscopic content analysis using salient information, dense disparity maps, and modified random walk framework

Stereoscopic content analysis tools are needed in 3D movie production to create exciting 3D effects while maintaining the visual comfort of viewers. Working practices in 3D movie production show that the behavior of human perception plays an important rule for determining the degree of comfort of a 3D content. In this paper, we propose perceptual based methods for stereoscopic content analysis. A modified random walk framework is proposed for dense disparity map estimation. A static analyzer that is useful for distinguishing blurred objects from clear ones is used for analyzing static visual importance. And a dynamic salient analyzer that is useful for distinguishing objects with single move or repeat moves in video frames is used for analyzing dynamic visual importance. These methods give us the flexibility to combine human perceptual factors for stereoscopic content analysis.