1992 Fiscal Year Final Research Report Summary
Study on Two-Dimensional Pattern Matching Algorithms
| Project/Area Number |
03650290
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
情報工学
|
|---|
| Research Institution | Ibaraki University |
|---|
Principal Investigator |
TAKAOKA Tadao Professor, Department of Computer and Information Sciences, School of Engineering, Ibaraki University, 工学部, 教授 (00007824)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
SAKATA Masateru Senior Engineer, System Laboratory, Hitachi Engineering Co., Ltd., システム研究所, 主任技師
|
|---|
| Project Period (FY) |
1991 – 1992
|
| Keywords | Pattern matching / Algorithms / Normalized cross-correlation / Sorting / Data compression / Color graphics |
|---|
| Research Abstract |
As computers are becoming multi-media devices, it is becoming important to speed up picture processing. In this research project, we pursue the speed-up of pattern matching algorithms as they are most fundamental in picture processing.
In the first year, research was done on the speed-up of pattern matching for texts and patterns whose pixels are given by 256-level gray values. The candidate points for matching is narrowed by using compressed texts and patterns where 8 X 8 segments are compressed into single values which are the means of gray values in the segments. That is, candidate points are obtained by the so-called 2-level pyramid method and then exhaustive search is done over near-by points by using the normalized cross-correlation function. For the matching in the compressed text and pattern we use the method that we sort the segments of the given text with the mean values and give gains to the candidates whose mean values are close to the mean values of the segments of the give
… More
n pattern. Our software for pattern matching with these techniques finds a 128 X 128 pattern in a 512 X 512 text in 3 seconds on a 12MIPS SUN workstation.
In the second year, a further speed-up was achieved. Specifically, the compressed text is made of partial samples of the text and non-promising candidates for matching are removed from consideration, saving much computing time. By this method, the computing time for the same setting is reduced to one second without losing reliability.
Furthermore in this year, research was done on the pattern matching for colored texts and patterns. Using the information of three colors, we can do pattern matching more reliably even for a little distorted patterns. Also a theoretical analysis was done on the approximate pattern matching with cappa mismatches for binary texts and patterns whose pixel values are 0 or 1.
The above mentioned results will contribute very much to picture processing algorithms. By the way, some results are achieved in graph algorithms in relation to graphics. Less
|
