Speaker
Description
Binarization of degraded documents is one of the pre-processing steps in character recognition tasks. The main task is to extract the text from the background and from other non-dominant texts that are common in this type of document. In recent years, methods based on deep learning have received more attention than traditional methods. Such an approach based on deep learning techniques requires large resources for training and the time required is considerable. In this work, we propose a hybrid model between a classical binarization method based on a parameterized 3D surface classification function and a deep learning approach. In this way, we reduce the complexity of the optimization model and the time required is significantly lower compared to the standard deep learning approach.
