Efficient motion vector recovery algorithm for H.264 using directional interpolation

This study proposes a novel motion vector recovery (MVR) algorithm for the H.264 video coding standard, which takes into account the change in the motion vectors (MVs) in different directions. Existing algorithms for MVR are confined to use the horizontal or vertical directions to recover the lost MVs. However, in the presence of non-linear movements or a fast/sudden motion of any object in a scene of the given input video, the MVs recovered by these algorithms turn out to be inaccurate. The proposed directional interpolation-based technique can interpolate the MVs in any direction based on the tendency of motion around the lost macro block, thus making it suitable to handle non-linear or fast motions. Testing the proposed technique with different benchmark video sequences shows an average improvement 1-2-dB in the peak signal-to-noise ratio of the recovered video over existing techniques.