On a method to decide kernel patterns of morphological associative memory

✍ Scribed by Toshihiro Ida; Syuji Ueda; Masayuki Kashima; Takayasu Fuchida; Sadayuki Murashima

• Publisher: John Wiley and Sons
• Year: 2002
• Tongue: English
• Weight: 298 KB
• Volume: 33
• Category: Article
• ISSN: 0882-1666
• DOI: 10.1002/scj.1144

✦ Synopsis

Abstract

Associative memory devices using computational morphology as proposed by Ritter are garnering a lot of attention. Ritter proposed two types of memory arrays, M and W, which have considerable levels of erosional noise and dilational noise, respectively. In addition, he has proposed an association method which combines the two memory arrays comprised of kernels when recalling a memory pattern. In this paper the authors investigate the recall rate by implementing the two‐stage recall method proposed by Ritter. However, this method cannot readily create kernel patterns when one memory pattern is included in other memory patterns or when a characteristic region for a memory pattern does not exist, and so the kernel pattern must be created by trial and error. Therefore, as the number of memory patterns increases, deciding the kernel pattern becomes more difficult. The authors demonstrate a method for creating kernels automatically for any memory pattern as a way to overcome this problem. It was found that in comparison to the associative memory device proposed by Ritter, the authors' resulting associative memory device could readily create kernel patterns for any memory pattern without relying on trial and error. © 2002 Wiley Periodicals, Inc. Syst Comp Jpn, 33(7): 85–94, 2002; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.1144