Student research opportunities

Image Completion/Inpainting---Removing Objects from Images

Project Code: CECS_1128

This project is available at the following levels:
CS single semester, Honours, Summer Scholar, Masters, PhD

Keywords:

computer vision, graphics, artificial intelligence, web

Supervisor:

Professor Stephen Gould

Outline:

Computer vision and computer graphics are, in some sense, inverse problems. In computer vision we aim to build a model from an image, in computer graphics we aim to produce realistic images from a model. Moreover, the fields share many algorithms in common. One area of overlap between the two fields is the problem of image completion or inpainting. Here the task is to complete a masked out region of an image to produce a realistic result. The masked region could contain an object (such as a person walking in front of a landmark), which you want removed, or some scratches or text that you want restored.

Goals of this project

The goal of the project is to develop a web-based system that integrates state-of-the-art image inpainting algorithms with human-assisted markup to produce high-quality image completions. Part of the project will involve recruiting Amazon Mechanical Turk workers to help evaluate/correct images.

The project may result in a scientific publication in a conference or journal.

Requirements/Prerequisites

Strong web-development skills on a Linux based system. Background in computer vision or graphics would be an advantage. A solid mathematics background and C/C++ or Python programming skills would allow additional emphasis on algorithm development.

Background Literature

* http://graphics.cs.cmu.edu/projects/scene-completion/
* http://research.microsoft.com/pubs/67276/criminisi_tip2004.pdf
* https://www.mturk.com/mturk/welcome

Contact: