著者

山口 高康 青野 博 本郷 節之

出版者

一般社団法人電子情報通信学会

雑誌

電子情報通信学会技術研究報告. PRMU, パターン認識・メディア理解 (ISSN:09135685)

巻号頁・発行日

vol.104, no.448, pp.1-6, 2004-11-12

被引用文献数

8 6

---

In this paper, we discuss image representation methods for signboard discrimination based on images. First, characteristics of image-based signboard discrimination problem and requirements for the discriminator are extracted. Next, traditional image representation methods, e.g. a color histogram or a space frequency, and so on, and their advantages and/or disadvantages are outlined. And next, effectiveness for each representation is compared for the signboard discrimination problem with images, using k-nearest neighbor (k-NN) method which is one of the most simple discrimination methods. Finally, to improve faculties of the image representation from the view point of signboard discrimination, two means, i) for reduction of the background affection and ii) for use of the color information, are discussed based on the extracted characteristics and requirements.