Robust motion estimation in video coding at low bit-rate
Hua Wang · 2004
A novel approach to optimize MC-DCT (motion compensated discrete cosine transform) hybrid video coding based on robust motion estimation is proposed in this paper. The approach introduces the robust estimation algorithm, i.e. least median squares (LMedS) method, from statistics into the block matching process for motion estimation in video coding. An important advantage of this approach is that it can result in better subjective video quality compared with traditional least mean squares (LMS) based block-matching algorithms, because robust motion estimation can detect the most significant underlying motion of the picture block more accurately, which can avoid the progressive degradation on subjective video quality due to the propagation of the errors owing to insufficient bits for coding the MC-DCT errors at low ing bit-rate. Experimental results obtained had supported our proposition on using LMedS-based matching criterion in block-matching motion estimation for low bit-rate video coding.