Abstract
A novel fast motion-estimation method based on an improved genetic algorithm is presented, in which both objective search and random search derived from genetic mutation are used for searching the global optimum and a threshold selection operator is applied to speeding up the estimation. The selection of initial population based on the coherence between neighboring macroblocks also improves the performance of search. Experimental results demonstrate that this method has very similar performance to that of FS, but just slightly slower than 3SS and 2DL. The inherent robustness and high parallelism enable it to be suitable for VLSI implementation of video encoders.
| Original language | English |
|---|---|
| Pages | 66-69 |
| Number of pages | 4 |
| State | Published - 1999 |
| Event | International Conference on Image Processing (ICIP'99) - Kobe, Jpn Duration: 24 Oct 1999 → 28 Oct 1999 |
Conference
| Conference | International Conference on Image Processing (ICIP'99) |
|---|---|
| City | Kobe, Jpn |
| Period | 24/10/99 → 28/10/99 |