A DNN Framework For Text Image Rectification From Planar Transformations

TL;DR

This paper introduces a neural network framework designed to correct distorted text images caused by planar transformations, demonstrating robustness and effectiveness in image rectification without explicit segmentation supervision.

Contribution

The paper presents a novel DNN architecture for text image rectification and provides a new dataset for evaluating such models.

Findings

The model can learn geometric transformations without explicit segmentation labels.

The proposed architecture effectively restores planar transformations.

The new dataset supports further research in text image rectification.

Abstract

In this paper, a novel neural network architecture is proposed attempting to rectify text images with mild assumptions. A new dataset of text images is collected to verify our model and open to public. We explored the capability of deep neural network in learning geometric transformation and found the model could segment the text image without explicit supervised segmentation information. Experiments show the architecture proposed can restore planar transformations with wonderful robustness and effectiveness.