Adaptive up-sampling method for scalable video coding of H.264

Abstract

In a scalable video coding (SVC), down and up sampling processes of video sequences are incorporated for spatial scalability. A new up-sampling method is proposed for H.264 SVC. The main idea is derived from the DCT-domain up-sampling scheme, which is efficiently derived from combination of the forward and backward DCT kernels. Two new up-sampling methods are proposed, which should be used appropriately in the encoder and decoder of SVC. First method is derived from the type-II DCT, which is a typical DCT for image compression. It suffers half-pel shift due to characteristics of the type-II DCT. The new up-sampling kernel is developed to support arbitrary-ratio up-sampling in the current H.264 SVC. Second method employs the type-I DCT filters, which have no phase shift. In order to remove the phase shift problem, encoder and decoder should use the same type-I DCT filters. Also, the developed up-sampling methods can use overlapping pixels near the block boundary to reduce the block artifacts and to improve performance. Fast algorithms are also developed by using various symmetries of the type-I and type-II DCT-based up-sampling kernels. Two methods are proposed for fast computation using direct and indirect matrix decomposition, where fast type-IV DCT is used in the indirect method. For performance improvement of the spatial scalability, an adaptive filtering method is introduced the up-sampling, which applies different weighting parameters to DCT kernels. In addition, an efficient estimation method of the weighting parameters is proposed by using the periodic characteristics of DCT. The proposed adaptive up-sampling method shows much improved PSNR in comparison with the recent SVC up-sampling method.