May 11, 2006
A Sufficient Condition for 1-D CNNs with Antisymmetric Templates to Perform Connected Component Detection
IEICE technical report
- ,
- ,
- Volume
- 106
- Number
- 30
- First page
- 15
- Last page
- 20
- Language
- Japanese
- Publishing type
- Publisher
- The Institute of Electronics, Information and Communication Engineers
We consider conditions for one-dimensional cellular neural networks (1-D CNNs) with the antisymmetric template A=[s,p,-s] to perform connected component detection (CCD). Given a 1-D black-and-white image, CCD is the task to obtain the output image satisfying 1) the number of black pixels is equal to the number of blocks of consecutive black pixels in the input image, 2) the right-most pixel is black unless all pixels are white, 3) all black pixels are isolated, and 4) there exists exactly one white pixel between two neiboring black pixels. Although it was shown experimentally by Matsumoto et al. that 1-D CNNs with the template A=[1,2,-1] can perform CCD, no theoretical proof has not been known so far. In this report, we analyze the dynamical behavior of 1-D CNNs with the antisymmetric template and give a sufficient condition for such CNNs to perform CCD.
- Link information
-
- CiNii Articles
- http://ci.nii.ac.jp/naid/110004745157
- CiNii Books
- http://ci.nii.ac.jp/ncid/AN10060800
- URL
- http://id.ndl.go.jp/bib/7938483
- ID information
-
- ISSN : 0913-5685
- CiNii Articles ID : 110004745157
- CiNii Books ID : AN10060800