High quality image resizing

• Authors: Qi Wang;Yuan Yuan
• Venue: Neurocomputing
• Year: 2014

An increasing amount of display devices with fixed sizes call for an adaptive strategy for optimal display. For this purpose, content aware image resizing techniques are developed. Previous works mainly lay their attention on the shrinkage operation of the examined image. Less efforts are paid on the expansion manipulation. Though some literatures claim an extension of their shrinkage operation to expanding the image in a similar way, the obtained results are not satisfying. In this paper, a high quality image resizing method is proposed to retain the details when stretching an image. Instead of using interpolation based techniques which are taken for granted by existing methods, an expansion model is first learned from a set of training images. Then the future enlargement is based on this principle. Experiments on two publicly available datasets demonstrate the effectiveness of the presented method. A further extension on video enlargement is also presented as an example. Though the proposed method is formulated in the context of seam carving, it can be readily extended to other techniques such as cropping, segmentation and warping based resizing methods.