Digital Image Processing Final Project

"Eye Finder in Digital Image"

Overview:

The final project will be organized as a competition and awards will be given to the top finishers. This project is designed to incorporate many of the teachings from this course (Digital Image Processing), however, students may incorporate learnings from outside of this course to solve the problem. E.g, Pattern recognition, symbolic artificial intelligence, artificial neural networks, etc.

There will be three portions to this project: 1) [25%] written report detailing the approach using the format provided (here), 2) [25%] an oral presentation of the system with examples of the system in use, and 3) [50%] the final competition which will be conducted by the instructor. Training examples are provided here.

Objective:

Locate eyes within a digital image that can contain a frontal face. The orientation of the faces will be mostly frontal (+10 to -10) degress of lateral rotation. All images will be 8-bit monochrome stored in the PGM format. The size of the images will vary as does the size of the eye(s) in the image.

Rules:

1. Students can form teams. Teams are limited to 2 students.
2. Teams will be provided a set of training images complete with ground truth data (manual located eye coordinates) in a text file. Test images will be provided for student development. The test images should not be used for training of the system.
3. Instructor will evaluate each teams algorithm with a sequestered set of images.

Evaluation:

The system will evaluated for correct eye match and location. Each eye match must return a processed image that placs a bounding box around the eye(s) and output in a text file eye locations in pixel coordinates (x,y). The pixel coordinates will be the center of the bounding box. The textfile will have the same filename as the input image except that the extension will be txt. Each eye found will correspond to a line in the text file of x-coord and y-coord followed by a newline character.

The system will be evaluated on how many eye images were located correctly (within 10x10 pixels) . The system will also compute the number of false matches, the number of eyes that were incorrectly located which includes when an eye was sad to be found although no eye was in the image.

Samples: