Differential Evolution Algorithm for Motion Estimation

Abstract

AbstractMotion Estimation (ME) is computationally expensive step in video encoding. Exhaustive search technique for ME yields maximum accuracy at the cost of highest execution time. To overcome the computational burden, many fast search algorithms are reported that limit the number of locations to be searched. ME is formulated as an optimization problem and the Sum of Absolute Difference (SAD) is considered as an objective function to be minimized. SAD error surface is a multimodal in nature. Fast searching algorithms converge to a minimal point rapidly but they may be trapped in local minima of SAD surface. This paper presents an application of Differential Evolution algorithm for motion estimation. The performance of the DE algorithm is compared with Full search, three step search, Diamond search and Particle swarm optimization for eight QCIF video sequences. Four performance indicators namely Peak Signal to Noise Ratio (PSNR), Structural Similarity (SSIM), number of search points and run time are considered for performance comparison of algorithms. Simulation result shows that both PSO and DE algorithms are performing close to Full search and reduces computational overload significantly in all the sequences.KeywordsBlock matchingMotion estimationDifferential evolutionParticle swarm optimization