Thesis Defense: Analysis and Transfer of Photographic Viewpoint and Appearance
Speaker: Soonmin Bae , MIT - EECS/CSAIL
Date: July 23 2009
Time: 2:00PM to 3:00PM
Location: 32-G449 (Patil/Kiva Seminar Room)
Host: Fredo Durand, MIT - Computer Graphics Group
Contact: Soonmin Bae, soonmin@csail.mit.edu
To make a compelling photograph, photographers need to carefully choose the subject and composition of a picture, to select a right lens and viewpoint, and to make great efforts with lighting and post-processing to arrange the tones and contrast. Unfortunately, such painstaking work and advanced skill is out of reach of casual photographers. In addition, for professional photographers, it is important to improve workflow efficiency.
The goal of our work is to allow users to achieve a particular photographic viewpoint and appearance with ease and speed. To this end, we propose to use model photographs. Given an input photo and a model photo, our algorithm analyzes viewpoint and appearance of the model photo and transfers the photographic characteristics to the input photo.
In this thesis,we present two contributions that transfer photographic view and look using model photographs and one contribution that magnifies existing defocus given a single photo. First, we address the challenge of viewpoint matching for rephotography. Our interactive, computer-vision-based technique helps users matching the viewpoint of a reference photograph at capture time. Next, we focus on the tonal aspects of photographic look using post-processing. Users just need to provide a pair of an input and a model photo, and our technique automatically transfers the look from the model to the input. Finally, we magnify defocus given a single image. We analyze the existing defocus in the input image and increase the amount of defocus present in out-of focus regions.
Our computational techniques understand users' goals and enhance their photographs by analyzing and transferring the photographic characteristics of model photographs. Our method increases users' performance and efficiency. We envision that this work would enable cameras and post-processing to embed more computation with a simple and intuitive interaction.
See other events that are part of
See other events happening in July 2009
