A Content-Aware Image Composition Approach Based on Random Walk

摘要

We present a novel content-aware image composition technique which can preserve the original color of the significant parts in the source image. By analysing the drawbacks of Poisson cloning, we propose our constraints the constructed membrane should follow. We also propose a modified random walk method which is able to meet these constraints with high efficiency and less user interaction. Two major advantages of our method are reflected. First, it provides a compact framework for combining the technique of image matting and image composition together, thus users need not pre-extract the object region where one want to preserve color intensity; second, it has the capability to compensate the artifacts which introduced by "abrupt color change" close to the image composition boundary. We carry on comparative experiments between our method and the up-to-date image composition methods to demonstrate that our method can achieve a more pleasing composition result with high efficiency.