Abstract: With the emergence of various smart devices, images often need to change the aspect ratio and sizes to adapt to different display screens. The content-aware image retargeting methods make the image automatically adapt to different display devices by researching how to keep image important content. This paper mainly summarizes the research status of content-based image retargeting methods in recent years. Firstly, the image retargeting methods based on handcrafted features are reviewed according to the acquisition of importance map and the retargeting method based on importance map. Secondly, the existing image retargeting methods based on deep learning are summarized from three categories：the image retargeting algorithm based on deep neural network, multi-operation algorithm based on deep reinforcement learning and cropping algorithm based on aesthetic perception. Thirdly, based on the introduction of the existing retargeting datasets and evaluation methods, the state-of-the-art image retargeting algorithms are analyzed and compared, and the implementation principles, advantages and disadvantages of each algorithm are summarized according to the categories. Finally, according to the existing problems and challenges at this stage, the future research direction in this field is put forward. This paper aims to provide interested researchers with meaningful help in order to promote the further development of this field.

Key words: image retargeting, deep learning, convolutional neural networks, retargeting datasets, evaluation metrics

摘要： 随着各种智能设备的不断涌现，同一图像往往需要改变纵横比和大小来适应不同显示屏幕。基于内容感知的图像重定向方法通过研究图像重要内容的保持，使得图像自动适应到不同大小的显示设备上。主要对近年来基于内容感知的图像重定向方法的研究现状进行总结。按照重要度图的获取以及基于重要度图的重定向方法两方面回顾了手工特征的图像重定向；从基于深度神经网络的重定向算法、基于深度强化学习的多操作算法、基于美学感知的裁剪算法三个类别总结了现有的图像重定向的深度学习方法；在介绍已有的重定向数据集以及评价方法基础上，对主流的图像重定向算法进行分析比较，并按照类别总结了各个算法的实现原理和优缺点；针对现阶段所存在的问题和挑战，提出了该领域未来的研究方向。该研究旨在为感兴趣的研究人员提供有意义的帮助，以便推动该领域的进一步发展。

关键词: 图像重定向, 深度学习, 卷积神经网络, 重定向数据集, 评价方法