Motion Vectors Estimation for Video Encoding Based on the Macroblock Signatures
Пыко Никита Сергеевич, Sergey Pyko, Oleg A. Mikus, Boris S. Filippov · 2020
To solve the problems of motion compensation during video encoding, algorithms are used that minimize the prediction error, based on either the sum of the absolute differences of pixels or frequency values after the Hadamar transform. The paper considers an alternative algorithm for finding the best approximation of the reference macroblock to the estimated one. For the search, the evaluated macroblock is presented in the form of a unique signature. To build a unique signature, the original macroblock of the video image is divided into microblocks. For each microblock, the coordinates of the center of mass are found. Depending on the required accuracy of the estimation of the motion vectors, the coordinates of the centers of mass are determined with an accuracy of a pixel, a half of pixel or a quarter of pixel. The set of coordinates of the positions of the microblocks mass centers defines an unique macroblock signature. As a criterion for the quality of the approximation of the estimated macroblock, the reference set is similar to the signatures: the mutual shift of the positions of the mass centers of the microblocks is minimized. The proposed algorithm is well vectorized and may be used for hierarchical estimation of motion vectors.