Saliency-Aware Image Completion

Abstract

In this paper, we present a saliency-aware image completion method which takes full advantage of a saliency detection result. On one hand, the saliency map is incorporated into the completion order computation procedure to take human visual attention into account. On the other hand, the saliency map used in patch matching makes the searched result more accurate and smooth. Furthermore, we employ an adaptive patch size determination algorithm which considers the color, structure, and saliency information simultaneously. Experiment results demonstrate the effectiveness of our system in preserving the structural information and robustness in various image content. We also show that the proposed system synthesizes more photo-realistic images than other image completion approaches.