| Presentation | 2008-02-22 A study for high performance character extraction from color scene images Masanori WAKABAYASHI, Keiichiro SHIRAI, Masayuki OKAMOTO, Hiroaki YAMAMOTO, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | This paper describes a robust method for extracting character strings from scene images. In our previous method, character patterns and the background ones are separated by clustering on the color space, then they are classified into character and background by SVM (Support Vector Machine). The extraction performance was not enough for the images where characters are arranged on a complex background, or the character color has changed by the influence of shading. The main reason of this is as follows. Clustering for a whole image region sometimes fails to separate the character color from the background color when their colors are similar or the size of character pattern is very small compared with the size of background pattern. To improve the clustering performance, previous method first divides the image into a rough block based on the edge, and applies clustering for each block. But in this block segmentation, when the background edges are very intense and connected to character edges, preferable block segmentation often can not be done. This paper describes a block segmentation method by using not only magnitude of the edge but also using color information on the edge. This method improves accuracy of the block segmentation, and as a result the performance of character extraction. Proposed method was tested with 251 images from ICDAR 2003 Text Locating Competition, and the effectiveness of this method is confirmed. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | character extraction / clustering / SVM |
| Paper # | PRMU2007-240 |
| Date of Issue |
| Conference Information | |
| Committee | PRMU |
|---|---|
| Conference Date | 2008/2/14(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Pattern Recognition and Media Understanding (PRMU) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | A study for high performance character extraction from color scene images |
| Sub Title (in English) | |
| Keyword(1) | character extraction |
| Keyword(2) | clustering |
| Keyword(3) | SVM |
| 1st Author's Name | Masanori WAKABAYASHI |
| 1st Author's Affiliation | Graduate School of Engineering, Shinshu University() |
| 2nd Author's Name | Keiichiro SHIRAI |
| 2nd Author's Affiliation | Graduate School of Engineering, Shinshu University |
| 3rd Author's Name | Masayuki OKAMOTO |
| 3rd Author's Affiliation | Graduate School of Engineering, Shinshu University |
| 4th Author's Name | Hiroaki YAMAMOTO |
| 4th Author's Affiliation | Graduate School of Engineering, Shinshu University |
| Date | 2008-02-22 |
| Paper # | PRMU2007-240 |
| Volume (vol) | vol.107 |
| Number (no) | 491 |
| Page | pp.pp.- |
| #Pages | 6 |
| Date of Issue | |