VEHICLE LICENSE NUMBER RECOGNITION SYSTEM USING NEURAL NETWORK

We developed equipment that recognizes license plate character for a parking lot. It is the character recognition equipment that can recognize all character on a license plate with applying a technique of image processing and neural network. In this paper, we propose a recognition way of the character that uses mosaic pattern as a feature value about recognition of KANJI and HIRAGANA on a license plate. License plate has various transformations on field data, for example out of position, tilt and so on. We focused low frequency spectrum in the image against these transformation plates, therefore using a rough mosaic as a matching pattern. Since a character on a license plate is few the number of dot per a character, a reducible ratio of input data and feature value is small. When we are about to make feature value, a big error has appeared on a resampling of an integral ratio. In order to settle these problems in a scarce processing time, we devised the way of making mosaic pattern. As a result, we came to be able to cut a high frequency element of pattern and normalize output data size at the same time to input various size data. We decide to call this technique expanded mosaic pattern method. From experiments with VCR, we confirmed the effectiveness of our proposed method, and can get classification rate of 99 percent on KANJI and HIRAGANA for each extracted character. (author abstract)

  • Supplemental Notes:
    • Five volumes of papers and one volume of abstracts comprise the published set of conference materials.
  • Corporate Authors:

    VERTIS

    TORANOMOM 34 MORI BUILDING 1-25-5
    TORANOMON, MINATOKU, TOKYO 105  Japan 
  • Authors:
    • Kojima, H
    • Yagi, M
    • SAKARI, K
    • Kurosaki, H
  • Conference:
  • Publication Date: 1995-11

Language

  • English

Media Info

  • Pagination: p. 174

Subject/Index Terms

Filing Info

  • Accession Number: 00719260
  • Record Type: Publication
  • Report/Paper Numbers: Volume 1
  • Files: TRIS
  • Created Date: Mar 5 1996 12:00AM