Computerized analysis of handwritten documents is an active research area in image analysis and computer vision. The goal is to create tools that can be available for use at university libraries and for researchers in the humanities. Working with large collections of handwritten documents is very time consuming and many old books and letters remain unread for centuries. Efficient computerized methods could help researchers in history, philology and computer linguistics to cost-effectively conduct a whole new type of research based on large collections of documents. The thesis makes a contribution to this area through the development of methods based on machine learning. The passage of time degrades historical documents. Humidity, stains, heat, mold and natural aging of the materials for hundreds of years make the documents increasingly difficult to interpret. The first half of the dissertation is therefore focused on cleaning the visual information in these documents by image segmentation methods based on energy minimization and machine learning. However, machine learning algorithms learn by imitating what is expected of them. One prerequisite for these methods to work is that ground truth is available. This causes a problem for historical documents because there is a shortage of experts who can help to interpret and interpret them. The second part of the thesis is therefore about automatically creating synthetic documents that are similar to handwritten historical documents. Because they are generated from a known text, they have a given facet. The visual content of the generated historical documents includes variation in the writing style and also imitates degradation factors to make the images realistic. When machine learning is trained on synthetic images of handwritten text, with a known facet, in many cases they can even give an even better result for real historical documents.

The current approach for text binarization proposesa clustering algorithm as a preprocessing stage toan energy-based segmentation method. It uses a clusteringalgorithm to obtain a coarse estimate of the background (BG)and foreground (FG) pixels. These estimates are used as a priorfor the source and sink points of a graph cut implementation,which is used to efficiently find the minimum energy solution ofan objective function to separate the BG and FG. The binaryimage thus obtained is used to refine the edge map that guidesthe graph cut algorithm. A final binary image is obtained byonce again performing the graph cut guided by the refinededges on a Laplacian of the image.

Most data mining applications on collections of historical documents require binarization of the digitized images as a pre-processing step. Historical documents are often subjected to degradations such as parchment aging, smudges and bleed through from the other side. The text is sometimes printed, but more often handwritten. Mathematical modeling of appearance of the text, background and all kinds of degradations, is challenging. In the current work we try to tackle binarization as pixel classification problem. We first apply semantic segmentation, using fully convolutional neural networks. In order to improve the sharpness of the result, we then apply a graph cut algorithm. The labels from the semantic segmentation are used as approximate estimates of the text and background, with the probability map of background used for pruning the edges in the graph cut. The results obtained show significant improvement over the state of the art approach.

The current approach for text binarization proposes a clustering algorithm as a preprocessing stage to an energy-based segmentation method. It uses a clustering algorithm to obtain a coarse estimate of the background (BG) and foreground (FG) pixels. These estimates are usedas a prior for the source and sink points of a graph cut implementation, which is used to efficiently find the minimum energy solution of an objective function to separate the BG and FG. The binary image thus obtained is used to refine the edge map that guides the graph cut algorithm. A final binary image is obtained by once again performing the graph cut guided by the refined edges on Laplacian of the image.

Accurate volumetric change estimation of glioblastoma is very important for post-surgical treatment follow-up. In this paper, an interactive segmentation method was developed and evaluated with the aim to guide volumetric estimation of glioblastoma. U-Net based fully convolutional network is used for initial segmentation of glioblastoma from post contrast MR images. The max flow algorithm is applied on the probability map of U-Net to update the initial segmentation and the result is displayed to the user for interactive refinement. Network update is performed based on the corrected contour by considering patient specific learning to deal with large context variations among dierent images. The proposed method is evaluated on a clinical MR image databas eof 15 glioblastoma patients with longitudinal scan data. The experimental results depict an improvement of segmentation performance due to patient specific fine-tuning. The proposed method is computationally fast and efficient as compared to state-of-the-art interactive segmentation tools. This tool could be useful for post-surgical treatment follow-upwith minimal user intervention.