Combination of Auto-encoder architecture and super resolution for better segmentation of thinned and cursive handwritten documents

In neural networks an auto-encoder architecture has several applications such as image denoising, feature reduction, data compression, image colorization, dimensanality reduction, segmentation and so on. Super-resolution is used to upgrade the low resolution images into high resolution. In order to get a better result on segmentation of thinned hand written images, this paper proposes a method of combination of associative auto-encoder architecture and super resolution for pixel expansion. Experimental results show that the combination of proposed network and super resolution method accurately segments the thinned handwritten Arabic words.