JVET-P1008 CE5-related: On the design of CC-ALF [K. Misra, F. Bossen, A. Segall (Sharp Labs of America), N. Hu, J. Dong, V. Seregin, M. Karczewicz (Qualcomm), P. Onno, C. Gisquet, G. Laroche (Canon), J. Li, C.S. Lim, C.-W. Kuo (Panasonic), J. Nam, J. Choi, J. Lim, S. Kim (LGE)] [late]

This contribution proposes a design for the Cross Component Adaptive Loop Filter (CC-ALF). CC-ALF operates as part of the adaptive loop filter process and makes use of luma sample values to refine each chroma component. The tool is controlled by information in the bitstream, and this information includes both (a) filter coefficients for each chroma component and (b) CTU level selection of the filter. The filter coefficients are signalled in the APS, and the appropriate APS is referenced in the slice header. A maximum of four filters are allowed per chroma component, and each filter is a 3x4 diamond shaped filter containing 8 unique coefficients. Filter coefficient dynamic range is limited to 6-bit signed, i.e. [-32, 31]. Filtering along virtual boundary make use of symmetrical line selection to align with regular ALF filtering. The coding performance of the design is evaluated using the Common Test Conditions (CTC) as compared to VTM-6.0, and the YUV BD-Rate is reported to be -1.58%, -2.11%, -2.38% and -2.33% for the AI, RA, LDB, and LDP test conditions, where the BD-Rate is calculated using the combined YUV metric of AhG13. To limit the per-pixel multiplier to 16 (current ALF is 15), a restriction is incorporated that enables only one of either chroma ALF or CC-ALF for a chroma component of a CTU. The coding performance of the design is evaluated using the Common Test Conditions (CTC) as compared to VTM-6.0, and the YUV BD-Rate is reported to be -1.30%, X.XX%, X.XX% and X.XX% for the AI, RA, LDB, and LDP test conditions, where the BD-Rate is calculated using the combined YUV metric of AhG13. Note: X.XX = values not provided in contribution.

See further notes under section 11.6.