Depth image super resolution via multi-hypothesis estimation

The rapid development of three-dimensional (3D) imaging techniques has significantly increased the demand for high resolution (HR) depth video and images. Significant pixel deficiencies and too much noise can be seen in depth images especially taken from Kinect cameras. For this reason, usability in several computer vision applications is restricted. In the acquisition of HR depth images, in traditional super resolution (SR) methods, either high frequency information or information attained externally from a HR database is obtained. In this study, multi hypothesis estimation (MHE) method using high frequency information of depth images with low resolution (LR) depth imaged patches is proposed. In this method, a HR depth image is created on the basis of similarities between the patches of the LR depth image, linear combination of patches circumferentially surrounding each patch. Experimental studies and comparisons have shown that the SR method with MHE is successful in creating HR depth images.