Real-Time Hard and Soft Shadow Compensation with Scene Adaptive Patch Similarity Gradient Pairs

Shadows are crucial for enhancing realism and provide important visual cues for many applications. This can be beneficial for the purposes like revealing information about the object’s shape, orientation, size, position, intensity, physical characteristics of the screen and even about the light source, type and behavior of shadowed surfaces. In recent years, many interesting contributions have been made both for hard shadows and soft shadows. This research emphasizes an approach toward real-time shadow compensation for dark/thick/hard and shallow/thin/soft shadows of captured scenes. 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 improve the object detection and make scene visualization more plausible. This research examines the patches in shadow and non-shadow regions and makes the best similar patch pair. These pair characteristics are used to reconstruct both soft and hard shadow regions adaptively, where soft shadows use gradient information for compensation. But, the hard shadows do not have scene information below the shadow area which is filled with adaptive gradient patch in-painting technique using close neighboring information. This proposed hybrid methodology shows improvement in the overall image quality. The efficacy of the methodology is proved in terms of both qualitative and qualitative evaluations.