Photo-metric Ambient Occlusion Reconstruction From Sparsely Sampled Illumination

As an inherent physical attribute,ambient occlusion play an important role in object three-dimension reconstruction,intrinsic decomposition,itDefines the local visibility for every pixel in physical filed,more specifically,it estimates the max angel between the normal and the direction for light which is visible for a pixel.Therefore,acquiring ambient occlusion is one of popular subjects in computer vision field.Nowadays due to many challenges in real condition,there are few effective methods to get the ambient occlusion conveniently?In this paper,we present a method to obtain ambient occlusion in a fixed view from a stack of sparse illumination?According to the simple statistical model,we calculate the light intensity of each pixel under the condition of sparsely sampling light.Then we use the relative entropy to constraints the consistency of theoretical light intensity distribution and practical light intensity distribution.Ultimately we get the estimation of ambient occlusion by solving a nonlinear least squares equation.Compared to the past solution,our approach seems more simple and efficient.Without the need to establish any constraints and relaxed requirements of the light condition,our method is more practical.In the experiment part of this paper,we build a complete validation dataset due to the lack of uniform validation dataset.Our dataset include four different objects,every object has a number of pictures which is shot in control of light.The more detail about our dataset is described in experiment part of this paper.In order to illustrate the accuracy and efficiency of our method,we compare our result with state of the art method on Tentacle dataset and the MIT dataset,this conclusions will be described in detail and be proved by objective experiment data.