Reducing computational complexity of adaptive motion estimation through binary comparison

A new technique for lowering computational complexity of adaptive motion estimation is proposed. Existing adaptive algorithms require a multi-bit data comparison per iteration and therefore enforce many energy-consuming signal transitions during the computation. We propose to reduce the transitions by replacing the 8-bit comparisons through 1-bit binary operations. Experiments show that such a technique can shrink the computational complexity of adaptive motion estimation by 82% for some frames with 17% on average without any impact on picture quality and performance. The hardware implementation is simple yet very compact.