Sharpness transfer for high-quality image composition

As the process of pasting a patch of interest from a source image into a target image, high-quality image composition always focuses on reducing the appearance gaps (boundary, color, brightness, sharpness, etc.) between source patch and target image. At present, in respect of sharpness unification, we often employ image filters to adjust the sharpness characteristic of one image. However, in composition process, it takes too much time and energy to manually update parameters of filter and continuously verify sharpness adjustment. In this paper, we propose a novel sharpness transfer approach for high-quality image composition, which can impose the sharpness characteristic of target image on source patch automatically and reasonably. First of all, specify source and reference patches in source and target images. Subsequently, the sharpness characteristic of each patch can be measured efficiently using a hierarchical estimation model. Then, the gradient of source patch is iteratively transformed until its sharpness estimation can reach the reference patch's. Finally, guided by the new transformed gradient, a new source patch can be reconstructed by minimizing an energy function with the constraints both in image domain and gradient domain, and be seamlessly integrated into target image. Experimental results demonstrate the effectiveness of our sharpness transfer approach. © 2012 American Scientific Publishers.