|
|
|
Maximum-Likelihood Image Matching
Clark F. Olson JPL internal report D-20097, Jet Propulsion Laboratory, California Institute of Technology, 2001. Image matching applications such as tracking and stereo matching commonly use the sum-of-squared-difference (SSD) measure to determine the best match. However, this measure is sensitive to outliers and is not robust to template variations. Object recognition methods using bounded errors to match images share a related problem, since the sharp distinction between matched and unmatched edge pixels can cause brittleness and non-graceful degradation. In this paper, we develop a new probabilistic formulation for image matching in terms of maximum-likelihood estimation that is applied to both edge template matching and grey-level matching. This formulation generalizes previous edge matching methods based on distance transforms. A likelihood function is constructed using the probability density function of the distance from the pixels in the model template to the closest matching pixel in the search image (according to some metric). We extend a branch-and-bound search algorithm to efficiently locate likely template positions in an arbitrary pose space. The use of an accurate probability density function allows precise subpixel localization and uncertainty estimation to be performed. These techniques have been applied to a diverse set of applications, including object recognition and stereo matching. In addition, the uncertainty estimation techniques allow feature selection to be performed for feature tracking and stereo matching by choosing features that minimize the localization uncertainty. |
