Improving the visualisation of 3D textured models via shadow detection and removal

Note: We don't have the ability to review paper

PubDate: October 2017

Teams: National Technical University of Athens

Writers: Evangelos Maltezos ; Anastasios Doulamis ; Charalabos Ioannidis

PDF: Improving the visualisation of 3D textured models via shadow detection and removal


Although shadows in images have a constructive role providing a natural view of features of the scene, they also have a destructive role in image processing by hiding significant information. Improving the quality of 3D textured models for serious games and augmented reality applications via shadow detection and removal remains challenging due to the complexity of an image scene. This paper proposes an efficient shadow detection and removal framework based on deep Convolutional Neural Networks (CNNs) and enhanced morphological operations. An orthoimage generated from high resolution RGB/UAV images, and a dense image matching digital surface model of a complex urban area of Santorini island in Greece, was used as a test dataset. First, the orthoimage is converted into the invariant to shadow color space of HSV. Then, a binary shadow mask is extracted applying the CNN which is trained using the HSV image and training samples polygons of shadows and nonshadow areas. To recover the shadowed areas of the orthoimage, a shadow removal process is applied via enhanced morphological operations. Finally, the corresponding 3D textured model colored by the final recovered orthoimage is extracted. The shadow detection and removal results illustrate the robustness, efficiency and the flexibility of the proposed framework.