Enhancement Of Text Recognition In Scene Images | ||||
| Mansoura Journal for Computer and Information Sciences | ||||
| Volume 13, Issue 1, December 2017, Page 19-26 PDF (830.27 K) | ||||
| Document Type: Original Research Articles. | ||||
| DOI: 10.21608/mjcis.2017.311954 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
| Moayed Hamad1; Osama Abu-Elnasr1; Sherif Barakat2 | ||||
| 1Faculty of computers and information systems , C.S dep. Mansoura University, Egypt | ||||
| 2Faculty of computers and information systems , I.S dep. Mansoura University, Egypt | ||||
| Abstract | ||||
| Text detection and recognition in natural scene images has received significant attention in last years. However, it is still an unsolved problem, due to some difficulties such as some images may have complex background, low contrast, noise, and /or various orientation styles. Also, the texts in those images can be of different font types and sizes. These difficulties make the automatic text extraction and recognizing it very difficult. This paper proposes the implementation of an intelligent system for automatic detection of text from images and explains the system which extracts and recognizes text in natural scene images by using some text detection algorithms to enhance text recognition. The proposed system implements various algorithms, such as Maximally Stable Extremal Regions (MSER) algorithm to detect the regions in the image, Canny edges algorithm to enhance edge detection and Bounding Box algorithm to detect and segment area of interest. Once the text is extracted from the image, the recognition process is done using Optical Character Recognition (OCR). The proposed system has been evaluated using public datasets (ICDAR2003 and the experimental results have proved the robust performance of the proposed system | ||||
| Keywords | ||||
| Scene Text Detection; Maximally Stable Extremal Regions; Bounding Box; OCR; OCR Spell Checker | ||||
|
Statistics Article View: 24 PDF Download: 19 |
||||
