Back to Search
Document details
Non-CE3: CCLM table reduction and bit range control
Abstract
This contribution proposes a method to reduce the CCLM bit range and table size. It is asserted that the VTM3.0 CCLM has two problems: (i) that the current design requires the use of two, large tables and (ii) that the precision of the entries in these tables necessitates high bit-depth multiplication. Specifically, it is observed that the VTM-3.0 CCLM employs two tables, each with 512 elements and where each element is 17 bits. This table is designed to avoid a division operation during the derivation process. However, when the internal bit depth of the decoded sample values is 10 bits, it is asserted that this design results in a 28 bit multiplication and 37 bit accumulation for the gradient a and intercept b, respectively, in the CCLM algorithm. It is argued that it would be desirable to reduce the calculation bit-depth to fit within 16 bits, which would have a benefit of both reducing the complexity of the calculation as well as the reducing the memory required for storing any needed tables.
JVET-M0064 Non-CE3: CCLM table reduction and bit range control [Y. Yasugi, F. Bossen, E. Sasaki (Sharp)]
Decisions
JVET-M0064 Non-CE3: CCLM table reduction and bit range control [Y. Yasugi, F. Bossen, E. Sasaki (Sharp)]
Citation