AUTOMATIC EXTRACTION OF TEXT IN IMAGES

• Main Authors: Chih-Hao Chen, 陳志豪
• Other Authors: Jia-Ching Cheng
• Format: Others
• Language: en_US
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/51693396047378099777

碩士 === 大同大學 === 電機工程學系(所) === 92 === Text detection and extraction is an active research topic in recent years. It has many practical applications including multimedia systems, digital libraries, and geographical information systems. In this thesis, we present a system, based on some generic features of text including homogeneity, physical constraints, and dominant stroke orientations, to detect and extract overlay text in digital images. First, we adopt the characteristic of dominant stroke orientations to devise an edge detector suitable for detecting edges along these dominant orientations. Next, physical constraints are induced to design an edge enhancer capable of extracting dense clusters containing text edges. Then physical constraints are further imposed to devise an text region detector consisting of (i) cluster parsing and regrouping which organize text in suitable units, (ii) a skew rectification which aligns the text region horizontally , and (iii) a text region verification which filters out false alarms of non-text regions. Finally, a specialized filter based on the features of dominant stroke orientations and an optimal thresholding based on the property of homogeneity are adopted to extract text embedded in complex backgrounds. We have applied the proposed system to detect and extract text of different languages in images. Experimental results show that it is robust for contrast, font-size, font-color, and background complexity