Effect of Saturation Arithmetic on Sum of Absolute Difference (SAD) Computation in H.264 Venkata Suman Sanikommu ECE 734 Project Presentation

Motion Estimation Block matching between successive frames – video compression  Find the best matching block (Motion Vector) Motion vector will be used to reproduce the reference frame  Motion vectors are found by calculating minimum SAD

Motion Estimation – SAD SAD Computation  Compute (A i – B i ) for all 16 x 16 pixels in the two blocks A and B  Determine which A i – B i is less than zero and produce the absolute value in that case, else produce A i – B i  Perform accumulate operation to all 16x16 absolute values.

Saturation Arithmetic Overflowed values will be represented as maximum values  Unsigned: 00…0h, FF…Fh Example:  6234h + E123h => FFFFh (saturated)

Saturation Arithmetic – SAD [1] Use saturation arithmetic and limit the number of bits used to represent SAD values  Reduced computation complexity  Reduced bits for SAD Might affect the quality of block matching and thus motion estimation

Saturation Arithmetic – SAD [2] If min{SAD} is less than FF…Fh  Does not affect motion estimation If min{SAD} is greater than FF…Fh  Affects motion estimation Subset size for block matching increases  Increased encoded file size We have to randomly select one for motion estimation  Randomly selected block might not be a best match

Project Work 1) Modify H.264 SAD computation code for saturation arithmetic 2) Compare the performance of H.264 video coding for modular arithmetic and Saturation arithmetic for different number of bits. 3) What is the minimum number of bits required to successfully use saturation arithmetic? 4) How frequently does the SAD value saturate for a given number of bits to represent? 5) What is the effect of saturation on encoded file size?

Questions?