Spatial anisotropic interpolation approach for text removal from an image

Abstract

We propose a Spatial Anisotropic Interpolation (SAI) based, Design and Analysis of Computer Experiment (DACE) model for inpainting the gaps that are induced by the removal of text from images. The spatial correlation among the design data points is exploited, leading to a model which produces estimates with zero variance at all design points. Incorporating such a feature turns the model to serve as a surrogate for predicting the response at desired points where experiment is not carried out. This property has been tuned for the purpose of gap filling in images also called as Image Inpainting, while treating the pixel values as responses. The proposed methodology restores the structural as well as textural characteristics of input image. Experiments are carried out with this methodology and results are demonstrated using quality metrics such as SSIM and PSNR. © 2013 Springer-Verlag.