In the current specification (JCTVC-L1003_v34) in section 7.4.9.3, in the semantics of sao_offset_sign, a temporary variable, named offsetSign is defined that specifies the sign for each SAO class.

Given the relationship of offsetSign with class index i, and of offsetSign later being used in equation 7-58 where offsetSign should vary within the for loop according to the value of i, it seems more appropriate to maybe, instead of defining a single variable, to define offsetSign as an array. As it is, it seems that maybe equation 7.58 could be misinterpreted.

Maybe also an alternative, instead of using offsetSign could be to use an equation of the form:

============
SaoOffsetVal[ cIdx ][ rx ][ ry ][ 0 ] = 0
for( i = 0; i < 4; i++ )
SaoOffsetVal[ cIdx ][ rx ][ ry ][ i + 1 ] =

(1 - 2 * sao_offset_sign[ cIdx ][ rx ][ ry ][ i ] ) * sao_offset_abs[ cIdx ][ rx ][ ry ][ i ] << ( bitDepth − Min( bitDepth, 10 ) )

============

and alter the semantics of sao_offset_sign so as it is appropriate initialized/set for all SAO categories, i.e.:

==============

If SaoTypeIdx[ cIdx ][ rx ][ ry ] then,

sao_offset_sign[ cIdx ][ rx ][ ry ][ 0 ] = sao_offset_sign[ cIdx ][ rx ][ ry ][ 1 ] = 0
sao_offset_sign[ cIdx ][ rx ][ ry ][ 1 ] = sao_offset_sign[ cIdx ][ rx ][ ry ][ 3 ] = 1

otherwise

– If sao_merge_left_flag is equal to 1, sao_offset_sign[ cIdx ][ rx ][ ry ][ i ] is inferred to be equal to sao_offset_sign[ cIdx ][ rx − 1 ][ ry ][ i ].

– Otherwise, if sao_merge_up_flag is equal to 1, sao_offset_sign[ cIdx ][ rx ][ ry ][ i ] is inferred to be equal to sao_offset_sign[ cIdx ][ rx ][ ry − 1 ][ i ].

– Otherwise, sao_offset_sign[ cIdx ][ rx ][ ry ][ i ] is inferred to be equal 0.

===============

Kind regards,

Alexis