Optimizing the Deblocking Algorithm for H.264 Decoder ImplementationWhat is “deblocking” ?Technical ChallengesProjection on Convex Sets (POCS) based Iterative Algorithm [Zakhor, 1992]Weighted Sums of Symmetrically Aligned Pixels [Averbuch, Schclar and Donoho, 2005]Adaptive Deblocking Filter [List, Joch, Lainema, Bjontegaard and Karczewicz, 2003]Proposed WorksOptimizing the DeblockingAlgorithm for H.264 Decoder Implementation By Ken Kin-Hung [email protected] is “deblocking” ?• Remove the discontinuity across block boundaries resulted from block-based DCT and coarse quantization steps in low bit rate encoding.[List, Joch, Lainema, Bjøntegaard and Karczewicz, 2003]Technical Challenges• Remove blocking artifacts without affecting the true edges in the picture so that the overall sharpness of the image is maintained.– A low pass filter is an intuitive choice, but the picture is blurred in general.• Computationally intensive, which requires as much as 1/3 of resources of the decoder.• High implementation complexity with the proposed algorithms based on complex mathematical operations.Projection on Convex Sets (POCS) based Iterative Algorithm [Zakhor, 1992]• One projection is on the set of signals that are bandlimited in both horizontal and vertical directions. Another projection is on the set of quantization intervals of the DCT coefficients.• Iteratively mapping a pixel value back and forth between the two sets until the value is within both sets.• There is no guarantee in the time to convergence due to the iterative nature of the algorithm.Weighted Sums of Symmetrically Aligned Pixels [Averbuch, Schclar and Donoho, 2005]• A pixel’s value is recomputed with a weighted sum of itself and the other 3 pixel values which are symmetrically aligned with respect to block boundaries.• The weights can be linear or quadratic.• Every pixel of the picture is going through a filtering process.Adaptive Deblocking Filter [List, Joch, Lainema, Bjøntegaard and Karczewicz, 2003]• An edge is classified with a set of conditions on the data points normal to the edge.• Filters with different strength are applied to different class of edge. For example, strong filter is applied on smooth blocks.• The algorithm is adaptive because the conditions are based on the quantization parameters included in the relevant blocks.Proposed Works• A simple blocking artifact detection algorithm which is easier to be implemented.– Keeping power consumption minimization in mind.• An algorithm identifying edges which will not result in a great improvement after filtering, and skip the filtering.– Trading off quality for a lower computation complexity.• Implementing the new algorithm on an open source reference codec and comparing the quality results with some existing
View Full Document