查詢結果分析
來源資料
相關文獻
- 相機取像的影像文字擷取與切割
- 多重專家應用於線上手寫中文文字識別
- The Survey of On-Line Chinese Character Recognition
- Applying Hashing Search and Fuzzy Fault-Tolerant Algorithms for the Fast Recognition of Multi-Font Printed Chinese Characters
- A Highly Reliable Character Recognition Mechanism
- Character Segmentation in Machine Printed Chinese Characters
- 應用聯線手寫文字辨識之前處理信號方法
- 聯線手寫日本語文字辨識中之連寫字探討
- 談中文光學文字辨識應用
- Building a Web-based Chinese Information Retrieval System Using Optical Character Recognition Technology: Project Report for Heleni Linton Bursary 1998.
頁籤選單縮合
題 名 | 相機取像的影像文字擷取與切割=Camera-based Optical Character Extraction and Segmentation |
---|---|
作 者 | 呂信德; 溫敏淦; 范國清; 林志瑋; | 書刊名 | 聯大學報 |
卷 期 | 6:2 2009.12[民98.12] |
頁 次 | 頁369-397 |
分類號 | 312.84 |
關鍵詞 | 相機取像; 文字識別; 文字辨識; 文字切割; Web-cam based OCR; Character segmentation; |
語 文 | 中文(Chinese) |
中文摘要 | 文字識別(OCR, Optical character recognition)已發展為人機界面中最可行、最有效率的輸入方式之一,為了提升OCR的應用,以利用相機取像設備普及的現況,本篇提出一套文字處理系統,可在使用者利用行動裝置的取像設備擷取有興趣的文件影像後,對此影像進行前處理盡可能得消除外在因素帶來的影響,並且以bottom-up的方式來處理所找出的文字區塊,並將其建構成文字行。對於每個文字區塊,我們提出一個結合連字過濾器的文字切割系統,藉由分離出單一字元與連字,來提高切割的效能。在實驗部分,我們使用200萬畫素webcam拍攝50張1600×1200名片影像,當中有9500個字元與419個連字(1,050單一字元),總共10600個字元,在低解析度相機無特殊取像限制、文字影像品質不佳的情形下,文字辨識系統的Recall rate仍可達到94.90%。 |
英文摘要 | Character segmentation is a critical problem in optical character recognition (OCR). Since images were captured from cameras, poor character images with the low resolutions generate the gluey and broken strokes. These problems increase the difficulty for segmenting the character images. In this paper, an image preprocessing technique is proposed to reduce the unwanted effects. Basically, it is bottom-up-based approach. Text line and typo-line data are first obtained from the connected components. Then, a character filter is performed to classify the characters into the single-character, the gluey-character, or the broken-character types. An algorithm is designed to split and merge the gluey and broken characters, respectively. Some experimental results demonstrate the feasibility of the proposed method with a recall rate 94.9%. |
本系統中英文摘要資訊取自各篇刊載內容。