A novel approach of generating stereoscopic images using defocus

Direct 2D-To-3D conversion is an important task with various manual or automatic methods to satisfy the rapidly increasing requirements on stereoscopic contents. In order to construct stereoscopic images, defocus is adopted to estimate the depth from a single image and a novel layered structure is proposed in this paper. A simple searching and filtering-based method is proposed for finding the best layering result with a comparison of performances between our method and certain existing unsupervised learning methods to determine the optimum approach for automatic choice of the layered threshold, which can be used to generate a set of layered RGB images by combining with the depth map. Then the planar layered model is constructed and texture is mapped from image onto the planes. With the help of the proposed approach, a layered model can be created from a single defocused image to provide stereoscopic feelings. The proposed approach has also been expanded to construct layered stereoscopic panorama, which shows its great application potentials.