The acceleration of fragment-based image completion using Gaussian pyramids

Abstract

We propose an acceleration method for fragment-based image completion [Drori et al. 2003]. The fragment-based image completion can synthesize a visually plausible image. However the time complexity of this method is O(n^2). We exploit Gaussian pyramids to improve and accelerate the algorithm. Moreover, to prevent the deterioration of resulting images caused by Gaussian pyramids, texture classification maps are introduced for specifying similar texture regions. The combination of Gaussian pyramids and texture classification maps allows fast image completion while preserving visual plausibility. This method can synthesize a fine image, e.g., one Mega pixel, within a minute.