EAGER: Aesthetically Empowering Novice Photographers

数码摄影和互联网使普通人能够以微不足道的可变成本拍摄和保存照片。因此，业余爱好者往往会积累大量的图像。然后，他们面临着从海量收藏中挑选出最美观的照片的繁琐琐事。这项任务的繁重性质因许多人的感觉而加剧，即阐明和量化美学属性是深奥的。根据用户定义的标准对照片进行自动评估还处于初级阶段，缺乏灵活性，无法提供针对个人用户的个性化体验。目前的系统基于图像数据集开发一组分类器，但这种方法阻碍了对用户偏好的适应，并抑制了开发新的美学启发式所需的灵活性。在这个探索性项目中，PI旨在通过创建软件来自动分类和过滤图像集合，从而提高业余摄影的整体美学质量，使用一套分析构建的美学原则，类似于专业摄影师所做的事情。该方法基于分层模块化分类器的概念，该分类器可以针对不同的个人进行定制，以便每个用户都可以拥有表征他或她的偏好的分类器。由于分类器是一个简单的模块排列，可以想象，即使是一个用户也可以设计多个分类器来满足他或她的每一个偏好；例如，用户可以有一个分类器来帮助选择S拍摄的最佳风景照片，而另一个分类器则用于选择最佳肖像。更广泛的影响：不同人口统计数据的比较将增强我们对文化、社会或其他背景因素如何影响个人审美偏好的理解。调整图像搜索以根据特定标准返回更美观的图像的能力将具有商业意义(例如，当公司努力创建针对不同受众的设计、产品或服务时)。这项工作也有社交网络应用，因为开发的算法可以让人们找到其他有相似偏好的人。PI的系统将教业余摄影师如何提高他们的拍照技能，向他们展示可以应用于增强照片的具体可能的修改。项目成果将为图像自动后处理技术奠定基础，从而提高图像的审美质量。因此，这项研究最终将提高网络上可用图像的整体质量。

Digital photography and the Internet are enabling ordinary people to take and save photographs at negligible variable cost. As a result amateurs often accumulate large volumes of images. They are then confronted with the tedious chore of sorting through their massive collections to select the most aesthetically pleasing pictures. The onerous nature of this task is exacerbated by the sense of many people that articulating and quantifying aesthetic attributes is esoteric. Automated evaluation of photographs based on user-defined criteria is in a nascent stage, lacking the flexibility to provide a personalized experience tailored to the individual user. Current systems develop a set of classifiers based on an image dataset, but this approach impedes adaptation to user preferences and inhibits the flexibility needed to develop new heuristics for aesthetics. In this exploratory project the PI aims to raise the overall aesthetic quality of amateur photography by creating software to automatically sort and filter a collection of images using a set of analytically-constructed aesthetic principles, in a manner analogous to what is done by a professional photographer. The approach is based upon the concept of a hierarchal modular classifier that can be customized for different individuals, so that each user can have a classifier that characterizes his or her preferences. Since the classifier is a simple arrangement of modules, one can imagine that even a single user could design multiple classifiers that cater to his or her every preference; for example, the user could have a classifier to help select the best landscape photographs s/he has taken, and another to select the best portraits. Broader Impacts: Comparison of data taken from different demographics will enhance our understanding of how cultural, social, or other background factors influence individual preferences in aesthetics. The ability to tune image searches so as to return images that are more aesthetically pleasing according to specific criteria will have commercial implications (e.g., when companies endeavor to create designs, products, or services that are tailored to different audiences). The work has social networking applications as well, since the algorithms developed could enable people to find others with similar preferences. The PI's system will teach amateur photographers how to improve their picture-taking skills, by showing them specific possible modifications that could be applied to enhance their photographs. Project outcomes will lay the foundation for technology that can automatically post-process images so as to improve their aesthetic quality. Thus, this research ultimately will improve the overall quality of the images that are available on the Web.