Image retargeting using nonparametric semantic segmentation

In this paper, a new full-automatic approach to content aware image retargeting is proposed. Most image retargeting approaches does not incorporate content information and only use local appearance information. However, there are some approaches which use high level information such as saliency regions, objects mask and depth information. Such methods do not use semantic labelling for each object. In this paper, object masks as well as their semantic class labels are used to propose a new approach to image retargeting. To do so, semantic segmentation of image is provided. Hence, a nonparametric approach to semantic segmentation is employed which is fast with no need to any learning model. This makes it simple and applicable to any dataset. To evaluate the proposed approach, besides presenting visual examples, we performed a set of subjective evaluations too. The obtained results show that our method outperforms other retargeting approaches.