Document Image Analysis Techniques for Handwritten Text Segmentation, Document Image Rectification and Digital Collation

Document image analysis comprises all the algorithms and techniques that are utilized to convert an image of a document to a computer readable description. In this work we focus on three such techniques, namely (1) Handwritten text segmentation (2) Document image rectification and (3) Digital Collation. Offline handwritten text recognition is a very challenging problem. Aside from the large variation of different handwriting styles, neighboring characters within a word are usually connected, and we may need to segment a word into individual characters for accurate character recognition. Many existing methods achieve text segmentation by evaluating the local stroke geometry and imposing constraints on the size of each resulting character, such as the character width, height and aspect ratio. These constraints are well suited for printed texts, but may not hold for handwritten texts. Other methods apply holistic approach by using a set of lexicons to guide and correct the segmentation and recognition. This approach may fail when the domain lexicon is insufficient. In the first part of this work, we present a new global non-holistic method for handwritten text segmentation, which does not make any limiting assumptions on the character size and the number of characters in a word. We conduct experiments on real images of handwritten texts taken from the IAM handwriting database and compare the performance of the presented method against an existing text segmentation algorithm that uses dynamic programming and achieve significant performance improvement. Digitization of document images using OCR based systems is adversely affected if the image of the document contains distortion (warping). Often, costly and precisely calibrated special hardware such as stereo cameras, laser scanners, etc. are used to infer the 3D model of the distorted image which is used to remove the distortion. Recent methods focus on creating a 3D shape model based on 2D distortion informa- tion obtained from the document image. The performance of these methods is highly dependent on estimating an accurate 2D distortion grid. These methods often affix the 2D distortion grid lines to the text line, and as such, may suffer in the presence of unreliable textual cues due to preprocessing steps such as binarization. In the domain of printed document images, the white space between the text lines carries as much information about the 2D distortion as the text lines themselves. Based on this intuitive idea, in the second part of our work we build a 2D distortion grid from white space lines, which can be used to rectify a printed document image by a dewarping algorithm. We compare our presented method against a state-of-the-art 2D distortion grid construction method and obtain better results. We also present qualitative and quantitative evaluations for the presented method. Collation of texts and images is an indispensable but labor-intensive step in the study of print materials. It is an often used methodology by textual scholars when the manuscript of the text does not exist. Although various methods and machines have been designed to assist in this labor, it still remains an expensive and time- consuming process, often requiring travel to distant repositories for the painstaking visual examination of multiple original copies. Efforts to digitize collation have so far depended on first transcribing the texts to be compared, thus introducing into the process more labor and expense, and also more potential error. Digital collation will instead automate the first stages of collation directly from the document images of the original texts, thereby speeding the process of comparison. We describe such a novel framework for digital collation in the third part of this work and provide qualitative results