The perspective rectification for non-planar documents

• Main Authors: Tsung-Hsien Wu, 吳宗憲
• Other Authors: Kuo-Chin Fan
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/66491614538096707522

碩士 === 國立中央大學 === 資訊工程研究所 === 99 === Recently, digital cameras become a universal device due to its cost down. People can capture images at will in any time. In the past, text images can only be acquired by scanning documents using scanners. Currently, we can obtain the images of any kinds of objects by simply using cameras within second. Resulting from the influences of environments existing in our daily life while capturing the object images, some new research topics arise, such as the space with uneven light-illumination and scene with more than one plane. Those new problems will definitely affect the performance of OCR (optical character recognition) drastically. Instead of focusing on OCR study which can already correctly recognize a single character with very high recognition rate currently, we devoting ourselves on slicing and obtaining a character from the images captured under poor conditions. For instance, the difference in view angles and texts do not distribute on the same plane. In this thesis, the research focuses on rectifying the documents with perspective distortions caused by different view angles while capturing images and various planes that texts locate on the image. In our work, we specially focus on classifying and rectifying images resided on cylinders, cubes of non-coplanar, and those captured through non-vertical light axis lens. This study provides an effective way in splitting an image with different planes and rectifying the split regions. The effective method in rectifying documents is designed mainly by using image processing techniques, such as connected-component labeling for extracting image information, text line extraction water flow algorithm, and image plane analysis. The proposed method can rectify those common document images with perspective distortion caused by non-singular planes without needing the information of document border and typesetting. Experimental results verify the feasibility and validity of our proposed method.