Parametrizations of planar models for region-merging based lossy depth-map compression

We present efficient methods for parametrizing planar models, to be used for depth value reconstruction inside selected regions in a depth image. The optimal plane for each region is represented using its quantized heights at three pixel locations. The decoder uses the decoded quantized heights to approximately represent the optimal plane. The three pixel locations are selected so that the distortion due to the approximation of the plane over the region is minimized. The planar reconstructions are used in competition with the piecewise constant reconstruction at the regions obtained through a merging process, where the two regions to be merged are those ensuring the optimal slope in the rate-distortion curve. The lossy depth compression algorithm including the planar modeling obtains a significantly better rate-distortion performance than the algorithm that uses only constant regions, with improvements up to 8 dB.