키넥트로부터 획득된 깊이 영상의 확장을 위한 GPU 기반 채움 기법

GPU Based Filling Methods for Expansion of Depth Image Captured by Kinect

Three-dimensional(3D) display technique is widely used in our daily life. Especially, to product augmented game contents which can interact with users, it is necessary to get high quality resolution image data to reconstruct 3D model more exquisitely. In this paper, we tried to expand depth image captured by Kinect using various interpolation methods(nearest neighbor, bilinear, bicubic) to adapt it to the size of original Kinect color image. To measure the quality of expanded depth image compared to original depth image, we used PSNR(Peak Signal-to-noise ratio) index. Besides, we implemented GPU parallel processing algorithm with OpenCL to interpolate a large amount of image data rapidly. As a result of the experiment, a bicubic interpolation method made an accurate depth image although it had a long time.