图像编辑是一项对图像的内容、风格、尺寸等可视因素进行制作和调整的技术，在平面设计、影视制作等领域有着广泛的应用。本项目力求对图像的可编辑性属性进行研究，基于互联网中海量的图像可视数据，利用深度神经网络提取多模态特征对图像在尺寸、色调风格以及内容等方面的可编辑性进行建模研究。项目拟解决的关键科学问题包括：数据驱动的图像可缩放属性的建模与预测，基于混合特征的图像可合成属性学习，基于多任务深度学习的图像可调色性模型构建。项目的创新性在于以数据驱动为主要核心机制对图像的编辑属性进行研究，同时从多种角度构造图像的特征表达，充分结合视觉认识机理和美学原理，致使图像编辑方法所生成的结果更加符合人类的审美和视觉倾向。

Image editing is a technology of making and adjusting visual elements in an image such as content, style, size and so on. It's popular and widely used in many fields, such as in graphic design and film productions. In this project, we aim to study the image editing properties with the help of massive visual media data from the Internet. More specifically, we plan to exact the multi-modal features from the internet images, and try to model the potential relationships between the extracted features and image editing properties. To implement this goal, deep neural networks are expected to be used in this project. The whole project includes three key scientific issues: data-driven based image retargetability modeling and prediction, hybrid-feature based image synthesizability learning and multi-task based image color adjustability modeling. The innovativeness exist in data-driven based image editing properties learning, multi-modal based image feature representation with combination of visual cognition mechanism and aesthetic principle. With above innovations, the project is expected to propose a series of image editing methods which can generate more satisfactory results in esthetics and visual tendency.