FlatDoc: Learning-Based Document Image Dewarping Using Geometric Rectification

Camera-captured document images often suffer from geometric distortions such as curvature and perspective skew due to non-planar surfaces and uncontrolled capture conditions.This paper presents a learning-based document image dewarping system that automatically rectifies distorted document images.The proposed approach combines document segmentation with geometry aware rectification by predicting a backward mapping grid and applying image warping to generate rectified outputs.The system is implemented as a webbased application that allows users to upload document images and obtain rectified results.Experimental results demonstrate effective reduction of geometric distortions and improved document readability.