Improved Depth Compression by Depth Downsampling Guided by Color Super-Pixel Refinement Segmentation
Mihail Georgiev, Atanas P. Gotchev · 2018
We propose an improved depth compression scheme which relies on depth decimation guided by super-pixel segmentation of the aligned color data. We modify the latter to ensure border congruency of segmenation refinement levels. Furthermore, a modification of our multi-modal regularized reconstruction is presented. We address also the problem of possible misalignments between color and depth maps. Such misalignments produce edge outliers which mislead the error optimization in the coding process. We propose an efficient encoding scheme of such outliers in so called “yieldflow” protocol. We compare our new and imporved method against a number of state-of-art approaches and demonstrate that it performs favorably especially in the low bit rate region.