A high quality hardware friendly motion estimation algorithm focusing in HD videos

Fast algorithms for motion estimation are often trapped in local minima, especially when working on high definition videos (HD). This paper presents a new algorithm for motion estimation focused on high definition videos named Multiple Iterative Random Search (MIRS). This algorithm uses randomness and multiple iterative steps as strategy to avoid local minima falls, achieving better quality results than traditional fast algorithms. MIRS is a hardware friendly algorithm since it has five iterative steps which did not have data dependencies, then these five iterative steps can be implemented in parallel, reaching a processing rate similar to other fast algorithms. That characteristic becomes MIRS a very competitive option for hardware implementation, since it is possible to reach very high processing rates with very good quality results. The comparative results show that MIRS algorithm presented the best PSNR among all evaluated fast algorithms. MIRS is also able to reduce in 70 times the number of evaluated blocks when compared with Full Search algorithm with a PSNR drop of only 0.71dB.