A new motion estimation algorithm based on adaptive hybrid genetic search and selective vector tracing

This paper describes a new motion estimation search strategy that exploits spatio-temporal correlation of a motion vector field to estimate any kind of sequences containing low/high spatial details and slow/fast moving objects. The proposed algorithm dynamically changes its searching strategy between a modified genetic search and an N-step search accordingly to the scene motion complexity and the specified processing power budget. An adaptive subsampled motion vector tracing among B-frames mechanism is provided that drastically improves the coding quality for a relative additional complexity (less than 25%) which gives better results than exhaustive search over large searching range in all CCIR601 sequences tested