Real-time hard and soft shadow compensation with adaptive patch gradient Pairs

Abstract

This research emphasizes an approach toward real-time shadow compensation for dark/thick/hard and shallow/thin/soft shadows of captured scenes. While humans are very good at estimating objects size, position, color, environmental changes, movements irrespective of occlusions and noise and hence, are able to smoothly visualize the scene. But, computing machines often lack the ability to sense their environment, in a manner comparable to humans. This discrepancy prevents the automation of certain real-time jobs and the shadows make it more cumbersome. Therefore, enhancement of shadow detected region patches with suitable compensation might change object detection and scene visualization more plausible. The authors examine the patch in shadow and non-shadow regions and make the best similar patch pair. These pair characteristics are used to reconstruct both soft and hard shadow regions. However, the hard shadows do not have scene information below the shadow area, that is filled with adaptive gradient patch in-painting technique using close neighboring information. This proposed hybrid framework shows improvement in the overall image quality in terms of both qualitative and qualitative evaluations. � Springer International Publishing Switzerland 2014.