Improvement of motion estimation accuracy using gradient vector distribution in advanced picture coding

This paper describes methods of improving the accuracy of a gradient-based motion estimation method that uses an eigenvalue algorithm. First, we propose reliability indexes of the estimated motion vector. Next, we classify the motion images by the proposed indexes into categories and analyze the characteristics of each classified image. According to the categories of the image characteristics, we use different methods in order to improve the accuracy of each motion estimation. Simulation results show that these proposed methods are effective in improving the motion estimation accuracy