Detection of textual information using convolutional neural networks

The low quality of scanned documents prevents one from recognizing the contents of them. However, often there are only a few important words that need to be distinguished. Spotting these words is a complicated task, which can be accomplished by utilizing the underlying document structure. The structure of the documents enables one to tackle the problem with the help of machine learning. This thesis investigates the boundaries of the object detection problem; however, unlike the previous research, we apply the detection pipeline to discover the words in structured documents.