Multiview image coding scheme transformations: artifact characteristics and effects on perceived 3D quality

Compression schemes for 3D images and -video gain much of their efficiency from transformations that convert the signal into forms suitable for quantization. The achieved compression efficiency is principally determined by rate-distortion analysis using objective quality evaluation metrics. In 2D, quality evaluation metrics operating in the pixel domain implicitly assumes an ideal display modelled as a unity transformation. Similar simplifications are not feasible in 3D analysis and different coding schemes introduce significantly different compression artefacts even though operating at the same rate-distortion ratio.   In this paper we have performed a subjective assessment of the quality of compressed 3D images presented on an autostereoscopic display. In the qualitative part of the assessment different properties of the induced coding artefacts was identified with respect to image depth, pixelation, and zero-parallax distortion. The quantitative part was conducted using a group of non-expert observers that assessed the 3D quality. In the results we show how the compression schemes introduce specific groups of artefacts manifesting with significantly different characteristics. In addition, each characteristic is derived from the transformation domains and the relationships between coding scheme and distortion property are presented. Moreover, the characteristics are related to the image quality assessment produced by the observation group