A combined DCT / DWT asymmetric multi-view video coding for real-time applications
The proposed 3-D DWT codec is 3-4 times less complex than x264 for the same objective quality. However, artifacts introduced by the wavelet transform, especially ''ghosting'' effects at moving objects can be more annoying to the subjects compared to traditional blocking and motion compensation artifacts introduced by the x264, even if the PSNR value achieved by the 3-D DWT codec is higher.
In order to overcome this problem we propose a new type of asymmetric multi-view video coding, when x264 is used for even views, while 3-D DWT is used for odd views in each stereo pair. The motivation behind this approach is the fact that the block-based motion estimation in x264 provides better visual quality for fast moving objects while wavelet scheme is free from blocking artifacts and provides better quality for the background (see the anaglyph video below).
During the subjective quality assessment test we found that the subjective quality of the proposed asymmetric scheme is always higher than the average sum of quality for 3-D DWT and x264. Herewith, the complexity of the asymmetric scheme is about 35% less than that of x264 codec.
Therefore, the DCT/DWT asymmetric multi-view video coding can be more preferable in many applications where real-time multi-view video capturing, encoding, and storage or transmission are needed.
 E.Belyaev, M.Georgiev, K.Egiazarian, and M.Gabbouj, A combined DCT/DWT asymmetric multi-view video coding for real-time applications // Eighth International Workshop on Video Processing and Quality Metrics for Consumer Electronics, 2014. [download]