A modification of the conjugate direction method for motion estimation
A comparative study of different block matching alternatives for motion estimation is presented. The study is focused on computational burden and objective measures on the accuracy of prediction. Together with existing algorithms several new variations have been tested. An interesting modification of the conjugate direction method previously related in literature is reported. This new algorithm shows a good trade-off between computational complexity and accuracy of motion vector estimation. Computational complexity is evaluated using a sequence of artificial images designed to incorporate a great variety of motion vectors. The performance of block matching methods has been measured in terms of the entropy in the error signal between the motion compensated and the original frames.