Authors: Deepak Kumar and A. G. Ramakrishnan

Description: After correcting the word images by unwrapping, word images are preprocessed by scaling the height. Images with height less than 60 pixels are scale up by three time the original height, images with height more than 180 pixels are scaled to a standard height of 180 pixels and remaining images are unaltered.
Middle row pixels of a gray scale image is segmented, using minimum and maximum values in a window, into two class [1]. Mean and variance of the classes are used to segment non-middle row pixels with neighborhood criteria.
Minimum value of image height and width, is used to pad zeros around the segmented image. Zeros padded images are passed to Omnipage OCR for word recognition.
[1] D. Kumar, M.N. Anil Prasad and A.G. Ramakrishnan, “MAPS: Midline analysis and propagation of segmentation,” Proc. 8Th ICVGIP, (2012).