A fast seam carving method based on merging seams in subimages

Abstract

Display devices based on its different screens and resolutions, need image resizing to retain the image's quality. Common image resizing methods cannot save or protect important objects or its results are non-photorealistic. Seam carving as a new method has been widely used for content-aware image and video resizing with little distortion in comparison with common methods. Unfortunately, seam carving is a complex algorithm and for high resolution videos or images has a long run time and is not usable in real-time applications. In this paper, a novel fast method in order to accelerate simple seam carving and decrease computational burden is presented. In this method image is divided into three equal horizontal or vertical sections, while the traditional seam carving is applied to the middle section. In the top and down sections, the algorithm estimates seam with respect to the middle part seam using an approximated Dijkstra method. Experiments have demonstrated better computational efficiency of presented method when it faces the current seam carving method. It is also preserving the image's information as effectively as the original seam carving method.