Depth images are frequently used in robotic and 3D vision for different purposes like mapping or object recognition. Yet recently, they are encountered in many other areas such as free viewpoint and 3D television. Their innate redundancy especially in high frame rates and resolutions demands an effective compression algorithm or otherwise the required data rates grow prohibitively large. Standard lossy image and video compression methods, such as JPEG2000 and H264, remove high frequency and usually unimportant components of the signal in intensity images; which in the case of range or depth images accounts for edges and are essential for correct reconstruction of the scene geometry. Therefore, preserving geometrical properties of depth images should be the main objective in an effective compression algorithm. In this paper, Wedgelets, Platelets and Wedge-Platelets are proposed for depth image compression and are compared with JPEG2000 and H264. Moreover, for the first time, these methods are applied for compression of Kinect sensor depth images. Compared with previous works, it is shown that higher compression ratios up to 10dB can be achieved.

Keywords