Content-aware media retargeting based on deep importance map

TL;DR

This paper introduces a neural network that generates importance maps for images and videos, improving content-aware media retargeting by reducing distortion and preserving significant regions more effectively.

Contribution

A novel neural network-based importance map estimation method that enhances media retargeting quality and handles difficult-to-resize images better than existing techniques.

Findings

Achieves better retargeting results with less distortion.

Performs well on enlarge operations and difficult-to-resize images.

Outperforms previous methods in objective evaluations.

Abstract

We present a neural network to estimate the visual information of important pixels in image and video, which is used in content-aware media retargeting applications. Existing techniques are successful in proposing retargeting methods. Yet, the serious distortion and the shrunk problem in the retargeted results still need to be investigated due to the limitations in the methods used to analyze visual attention. To accomplish this, we propose a network to define the importance map, which is sufficient to describe the energy of the significant regions in image/video. With this strategy, more ideal results are obtained from our system. Besides, the objective evaluation presented in this paper shows that our media retargeting system can achieve better and more plausible results than those of other works. In addition, our proposed importance map performs well in the enlarge operator and on the "difficult-to-resize" images.