An Adaptive Warp Correction Algorithm for Handwritten Text Images with Non-Linear Baselines

Skewed text can be explained as a highly variable spatial arrangement of accurately shaped letters with non-linear base. A poorly written text incurs the element of randomness into it. Offline detection of poorly written and skewed text is a continuing study in image analysis. The worst case scenario being the baselines arbitrarily warped without any correlation among them, making them difficult for existing computational methods to rectify automatically. The present work provides an Adaptive Warp Correction Algorithm for detection of skewed handwritten text using readily available programming tools for better readability, easier analysis, and cross-platform support. The algorithm proposed in this work utilizes extensive mathematical morphology with the help of adaptive kernel management based on the image dimensions and local region based data analysis to obtain an indexed map of the lines in the image. It then warps the skewed lines using piecewise affine transformation on a per-line basis to obtain the straightened image. The final image is obtained by stitching the straightened line images together.