Extended Adaptive Loop Filter Beyond VVC

Adaptive Loop Filter (ALF) has been adopted into VVC standard, which significantly improves the video compression performance. In this paper, we propose two extensions for ALF beyond VVC to further increase the coding efficiency. First, a cross filter shape with long filter length is proposed to utilize information from the reference samples far away from current filtering sample. Different from the existing diamond filter shape in VVC, the proposed new filter shape can further explore the correlations between long-distance samples. Second, more extended taps for ALF are introduced into online-trained-filters to involve more texture information provided by the offline-trained-filter-output of ALF. Compared with the existing design in VVC, more diversified information is input into ALF. The proposed extensions for ALF have been implemented on top of ECM software published by JVET. Experimental results show that the proposed extensions can achieve -0.13%, -0.22% and -0.41% coding performance on the luma component in All-Intra, Random-Access and Low-Delay-B configurations respectively, on top of ECM-5.0, with negligible encoding/decoding complexity impact. The proposed extensions have been both adopted into ECM.