Natural Scene Text Localization and Removal: Deep Learning and Navier-Stokes Inpainting Approach

Algorithm utilizes adaptive gaussian binary thresholding, text localization, dynamic masking and TELEA/Navier-Stokes Inpainting Algorithm. An algorithm is proposed that provides effective text removal from scenes, real world images containing text. It directly predicts and localizes scene text in any given image. A number of examples on real world images containing scene text demonstrate the effectiveness of the algorithm in removing large as well as smaller detectable texts which uses the ‘An Efficient and Accurate Scene Text Detector’ Deep Learning Model.