Area-Efficient Processing Elements-Based Adaptive Loop Filter Architecture With Optimized Memory for VVC

In-Loop filter tools are essential to video codecs as they bring significant compression performance. Versatile Video Coding (VVC) introduces a new filtering tool named adaptive loop filter (ALF), which brings over 4% performance improvement at the expense of high computational complexity. This brief presents a high-performance ALF architecture embedded with optimized memory and area-efficient processing elements (AE-PEs). Firstly, a limit scheme is proposed, where the filter coefficient values are analyzed and limited according to their concentration. Then, we propose two mapping modules to eliminate redundant look-up tables, saving memory resources during the coefficient derivation. To reduce the redundancy in filters, three AE-PEs with fewer adders are designed, whose datapath can be reconfigured by coefficient signals. According to the experimental results, the limit scheme brings a certain performance gain due to the better trade-off between the rate and distortion. The proposed ALF hardware architecture supports processing 4K@33fps videos and achieves more than 41% normalized area reduction than previous works.