PERSONAL PHOTO ENHANCEMENT

Abstract:

Thanks to the democratization of digital cameras and camera-equipped smartphones, casual photographers have an increasing need to be able to easily enhance their growing personal photo collections. Existing commercial tools for photo enhancement are either expressive but too sophisticated for casual photographers (e.g. Adobe Photoshop), or easy-to-use but limited in expressiveness (e.g. one-click filters from Instagram). This poses the challenge
of how to enable novices to easily edit and enhance their personal photo collections. This thesis describes three systems that collectively build a foundation for solving the practical problem of enhancing personal photo collections as a whole.

Our goal is to enhance a personal photo collection, and the first challenge is to select which photos in the collection are worth keeping, editing and sharing. In particular we would like to automatically exclude bad photos from a series of shots taken of the same scene. Note that this goal takes a different perspective from that of previous work, for two reasons. First, casual personal photos have different statistics from those of general images. Second, the problem of selecting which are the best photos in a series of similar shots is different from establishing an overall quality measure for any single photo in isolation.

Next, the thesis studies the color transfer problem on photo enhancement. Because visual taste is personal, an enhancement tool is necessary to provide users with both enough freedom and intuitive control. Therefore, additional to a novel color transfer algorithm, we also design a simple interface for making color exploration easy and intuitive. When taking a photo collection as a whole, the color transfer algorithm can also be applied to enhance the overall color consistency among multiple photos automatically, which also enables users to tune their entire photo album simultaneously.

Finally, we narrow down photo enhancement to specific domains and focus on
portraiture in personal photo collections. We propose a novel framework for modifying portrait images to match with the style of another person in a reference photo, and in particular, we explore the application of this technique towards transferring makeup styles.