A Novel Processing Chain for Shadow Detection and Pixel Restoration in High Resolution Satellite Images Using Image Imposing

High resolution satellite images may contain shadows due to the limitations of imaging circumstances and presence of tall-standing objects. These shadows cause problems in the exploitation of such images. This paper proposes a complete processing chain to mitigate these shadow effects. This processing chain has two parts. A shadow detection part bases on image imposing and a pixel restoration part based on Bayesian belief propagation algorithm. The shadow detection part executes a Binary conversion supervised by Support Vector Machine (SVM) algorithm and a Canny Edge Detector followed by Image Imposing. The pixel restoration part has two phases. An example-based learning phase and a conclusion phase. In the example-based learning phase, the positions of the shadow and nonshadow pixels are observed and stored in different libraries named as shadow and nonshadow library. By exploiting their Markov Property, the samples are connected by Markov Random Field (MRF). In the conclusion part, the relationship learnt from MRF is used for Decision Making by Bayesian Belief Propagation algorithm. In the end, the restored image is evaluated by comparing the Image Enhancement Factors (IEF) of the images which are shadow detected by image imposing and by morphological filtering. The results after the evaluation on the satellite images exhibit that the shadow detection part is fine and the recovered shadow regions are consistent with their neighboring nonshadow region. KeywordsShadow Detection, Pixel Restoration, Support Vector Machine (SVM), Canny Edge Detector, Image Imposing, Example-based learning, Markov Random field (MRF), Bayesian Belief Propagation (BBP), Image Enhancement Factor (IEF).