الفهرس 
Abstract
Problem DefinitionOnly 14 pages are availabe for public view
 |  | from | 16 |  |  |
| | | from | 16 | | |
Abstract
Pattern Matching is a fundamental problem in computer vision, and image and video processing, with a large number of applications that heavily relies on pattern matching as one of their building blocks. The central aspect of current research on pattern matching is achieving balance between efficiency and quality.
Pattern Matching is fundamentally performed in two ways: Either find only the candidate template that best matches the query template, or find all templates with distance to query below a certain threshold.
Recently, there have been great interest in fast, exhaustive search equivalent pattern matching algorithms. These algorithms provide results that are identical in terms of quality to exhaustive search, but significantly faster, their work is based on providing sufficient conditions for rejecting candidate patterns, so when these conditions are met, all computations associated with examining a candidate pattern can be saved.
We propose a novel exhaustive-search equivalent algorithm, called Norm Ordered Matching (NOM), which is combined with some other algorithms to provide significant speed-ups over current state of the art algorithms in the problem of finding nearest pattern. Our technique also shows high resilience to both blurring and JPEG compression types of noise. This is demonstrated with results from over 15 million runs for each compared algorithm. We also demonstrate the efficiency of our algorithms by integrating it in the exemplar image inpainting algorithm, and showing that this achieves drastic speed-ups without affecting the quality of the inpainting process.