Project Goal
Imagine a commercial camera which has a "smile" button: Whenever the button is pressed, all people in the image smile, no matter what their actual facial expression was then taking the image. Wouldn't our grandparents love such a camera!
Figure 1: Faces detected by online software (S. Baluja and H. Schneiderman, CMU)
Project Scope
You are asked to develop techniques for manipulating images so as to make all faces smile. This might involve using an online face detector, to segment out all faces. There should be a learned deformation between a normal face and a smiling face, that can be applied locally to morph a regular into a smiling face. Optimally, you might develop technique for classifying the amount of deformation necessary to make a face smile but still look natural, and you might learn additional modifications, such as adding mustaches to all people in the image.
Tasks
The project will be accomplished through the following tasks.
- Task 1: Implement an interface to an online face finding algorithm, or alternatively, implement a face finding algorithm. One candidate here CMU;s online face detector
- Task 2: Collect training images of faces with different degrees of smile.
- Task 3: Learn a transformation from regular to smiling faces using Markov random fields (which we'll tell you all about)
- Task 4: Apply the transformation under appropriate scale and orientation to faces in an image, where faces are detected and localized using the face segmentation algorithm.
- Task 5 (optional, but very cool); Develop a learning algorithm which estimates the degree as to which a person smiles already in an image, and use it to determine the degree at which a face has to be bent
- Task 6 (optional): Apply this algorithm to images, and generate an image of naturally looking smiling faces.
- Task 7 (optional): Do the same with mustaches.
Project Status
Doantam Phan, Ron Yeh, and Ajay Gupta
Point of Contact
Sebastian Thrun
Midterm Report
submitted
Final Report
submitted
