本项目将以网络海量可视媒体的特征分析和语义挖掘为基础，以数据驱动机制的为核心研究图像合成方法，实现图像内容的评价、优化和自动生成，探索利用网络可视媒体数据对图像合成相关理论和技术的智能化提升机制。本项目将首先研究基于视觉先验的可视媒体内容结构分析与学习方法，在面向内容生成的视觉信息表达机制和语义特征关联关系学习等方面进行研究。以此为基础，研究图像美化方法，探索图像色调美化和人像美化的数据驱动方法；研究图像缩放方法，探索图像可缩放性属性的学习和预测方法，构建数据驱动的图像缩放方法优化选择机制；研究图像风格化方法，在数据驱动的自然图像风格化、图像卡通化、国画绘制过程模拟和风格化生成方面进行探索。

This project will base on the feature analysis and semantics mining of internet visual media, and use data-drive mechanism as key method to study image synthesis methods, including assessment, optimization and automatic generation of image content. The goal is to explore the schemes of utilizing internet visual media to develop intelligent image synthesis theories and methods. We will first study the visual prior-based visual media content analysis and learning methods, study content generation-oriented visual information expression scheme and learning method of semantic feature association. On this basis, study image beautification methods, including data-driven methods of image color beautification and portrait beautification. Study image resizing methods, including the learning and predicting methods image resizability properties, and data-driven image resizing method selection method. Study image stylization methods, including data-driven methods of natural image stylization, image cartooning, simulation of Chinese painting drawing process and generation of Chinese painting from natural images.