A combined Convolutional Neural Network and Dynamic Programming approach for text line normalization

This work proposes a new normalization algorithm for handwritten text lines based on the use of Convolutional Neural Networks trained to classify pixels of the scanned text line as belonging to the main body area. The reference lines of the text line are obtained from these local estimates by means of Dynamic Programming. The obtained reference lines are used to normalize the text line images. Experimental results on the IAM offline database demonstrates the feasibility of this approach.