Reliability index of motion estimation based on eigenvalue algorithm

In this paper, we propose new reliability indices as an indicator of accuracy for the estimated motion vector. In gradient-based motion estimation with eigenvalue algorithm. the gradient vector distribution closely is related to estimated motion accuracy. We express the reliability indices using eigenvalues that represent the gradient vector distribution. Experimental results show that the proposed reliability indices are better indicators of the accuracy of motion estimation than motion-compensation prediction error. This paper describes also that the proposed reliability indices are useful to categorize the image in accordance with the main causes of erroneous estimation.