140 likes | 268 Views
This paper evaluates various search strategies for block matching in image processing. The main objectives include finding the best match for a given block while minimizing the number of block comparisons. It discusses the effectiveness of strategies such as full search, 2D logarithmic search, and conjugate direction search, analyzing their performance based on distortion measurement, mean square error, and test data from known sequences. The findings reveal significant speedups compared to full search methods, providing insights for practical applications in image compression and motion estimation.
E N D
EE368b Project Evaluation of search strategies for block matching Marcus Isaksson Joakim Jalden Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Block matching • Main goals of a block matching scheme • Find best match for any given block • Minimize number of block comparisons • Contradictory goals Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Presentation overview • Distortion function • Search strategies • Test data • Algorithm performance • Conclusions Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Distortion measurement • Mean square error • Bilinear interpolation is used for non-integer values of i and j. Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Search strategies • Full search • 2D logarithmic search • Jain J.R., Jain A.K., 1981 • Conjugate direction search • Srinivasan R., Rao K.R., 1985 Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Conjugate direction search Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Description of test data • Suzie1 • Frames 47-52 from the suzie qcif sequence • Limited motion in parts of the image, different directions • Foreman • Frames 308-313 from the foreman qcif sequence • Panning camera, large motion over the whole image Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Suzie1 Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Foreman Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Distortion function Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Choosing a start point for the search • Use motion vectors from adjacent blocks Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Results Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Results Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson
Conclusions • Major speedups compared to full search • Number of block comparisons reduced by a factor 100 for integer pel • 500 for quarter pel • Conjugate direction with memory is the fastest strategy, but sensitive to selection of start point • If time is not a problem: choose full search Evaluation of search strategies for block matching Joakim Jald₫n and Marcus Isaksson