This thesis is meant to investigate new multi-view-plus-depth (MVD) coding frameworks limiting as much as possible the perceptible distortions occurring in the views synthesized from decompressed data. Our objectives thus particularly concern the development of perceptually driven tools in the context of MVD coding. The difficulties we tackle lie in the fact that not only the distortions sources are multiple but there is no dedicated quality assessment tool for this specific type of data. For this reason, this thesis is focused on two main issues: the synthesized views quality evaluation and their specific artifacts and the conception of new compression scheme for MVD data based on perceived quality criteria. In this thesis, we conducted several studies in order to qualify DIBR related artifacts in synthesized views. Student's t-test results from ACR-HR and Paired Comparisons scores have led to the determination of reliability of usual subjective image/video quality assessment methods when used for the evaluation of synthesized views. The performances of usual objective image/video quality metrics has been evaluated through their correlations with subjective scores. We then focused on depth maps compression. We proposed two depth map coding schemes that are based on a compression method for 2D still images, namely LAR method. Based on our observations and experimentations, we proposed a representation and coding strategy that meets the need for depth edges preservation while ensuring high compression rate. Comparisons with state-of-the-art codecs such as H.264/AVC and HEVC showed improvements in terms of visual quality at low bit-rates. We also conducted studies regarding the bit-rate allocation issue in the context of MVD coding. The results of this thesis can be useful for the conception of new image/video quality assessment protocols, in the context of MVD; for the conception of new objective quality metrics; for improving MVD coding schemes by integrating our proposed tools; for the optimization of MVD coding methods from our studies on the relationships between texture and depth data.
