Image Retargeting Based On RGB-D Images

Image retargeting is a process changing the resolution of images to fit various displays. With the rapid development of multimedia, conventional image retargeting techniques can not satisfy the requirement today, and the content-aware image retargeting become an attractive topic in the research of computer vision. In this paper, we present an image retargeting method based on RGB-D images in the framework of seam carving. We modify the seam carving algorithm in the two aspects, image understanding and retargeting strategy, to improve the performance and robustness.The main contents of this paper are as follows:(1) Considering 2d features in RGB images, such as brightness, color and texture, can be easily influenced by the light and the complex background, we make a research on RGB-D images in this paper, and focus on the feature extraction in RGB-D images.(2) We present a self-adaptive energy model based on RGB-D images. The energy model unites gradient energy and self-adaptive saliency energy, which is based on the distribution of depth map. The performance in experiments prove our energy model has a strong robustness in various images.(3) We present a novel piecewise seam to improve the flexibility of seam and we modify the strategy in the optimal seam searching. Experiments illustrate that our method has a better performance under complex image texture.The work of this paper has a positive role in promoting content-aware retargeting performance and has a certain reference value for other related algorithms.