Fast wavelet-based multiresolution image registration on a multiprocessing digital signal processor
✍ Scribed by Hao Wu; Yongmin Kim
- Publisher
- John Wiley and Sons
- Year
- 1998
- Tongue
- English
- Weight
- 201 KB
- Volume
- 9
- Category
- Article
- ISSN
- 0899-9457
No coin nor oath required. For personal study only.
✦ Synopsis
Image registration is a fundamental task in image prothe images are most similar to each other. Most current research cessing. It is used in matching two or more images taken at different on image registration is focused on two aspects of this search times, from different imaging modalities, or from different viewpoints.
process: the similarity measure and the search strategy.
One of the obstacles in achieving practical acceptance of image regis-Among the various similarity measures that exist, the crosstration techniques is their computational complexity, which results in correlation coefficient is the most frequently used [1,2,4,7,13,14].
a long response time. In this article we present a fast multiresolution A far more computationally efficient similarity measure based on image registration algorithm using wavelet transform for the translathe sum of absolute differences between the pixels on the two tional and rotational alignment of two-dimensional images. In particuimages was proposed by Barnea and Silverman [22]. Venot and lar, a novel approach to determine the algorithm parameters to bal-
Leclerc [4] introduced two similarity criteria based on sign
ance the registration accuracy and computational requirement is also described. We implemented this algorithm on a PC-based multimedia changes: the deterministic sign change (DSC) criterion and the and imaging system using a multiprocessing digital signal processor.
stochastic sign change (SSC) criterion. Applications of these
The algorithm is capable of achieving a subpixel registration accuracy similarity measures could be found in their later work on photoreliably under various noise levels. The multiresolution algorithm imgraphic images [6] and medical imaging [10]. More recently, plemented on this desktop system was able to register two 256 1 Chiang and Sullivan [23] proposed a new similarity measure 256 images in 466 ms, which is 40 times faster than the uniresolution called coincident bit counting (CBC), which is based on the exhaustive search approach.