3 boyutlu videonun derinlik yönlendirmeli çözünürlük iyileştirmesi

This thesis investigates the problem of super-resolution (resolution enhancement) inmulti-view 3D video.Although the super-resolution problem has been widely studied for monocular video,work on 3D super-resolution of stereo/multi-view video is rather new. Given recentadvances in depth-based stereo/multi-view video representations and compression formats,this thesis proposes a depth-guided super-resolution method for multi-view 3D videos.Since sub-pixel registration of multiple views is crucial in achieving super-resolution, weemploy a parametric registration approach based on affine motion of approximately planarimage patches rather than traditional pixel-based stereo matching. The segmentation ofplanar image patches and their affine registration is guided by depth layer maps.We also conduct subjective visual evaluation tests of super-resolution in 3D on differentstereoscopic displays.