CS 269 Project Page

Fall 2014

Project Abstract

Automated handwriting detection remains an interesting yet challenging problem in the computer vision field. This project investigates an approach from Hinton et. al. that uses an elastic matching method to recognize digits. Each digit class is represented by a cubic b-spline in an "ideal" configuration. To classify a test image, an iterative algorithm performs an elastic match between the test image and each digit model. Validation is performed against the publically-available handwritten digit dataset, MNIST.

Group Members

Ansuya Ahluwalia, ansuya@cs.ucla.edu

Eric Kim, ekim@cs.ucla.edu

Nicholas Brett Marcott, bmarcott@ucla.edu

Project Materials

The project report PDF can be found here.

You can download the source code (Matlab) here.

Finally, you can view several animations that illustrate the digit fitting process:

Fitting models to images of "2"	Ex1, Ex2
Fitting models to images of "3"	Ex1, Ex2

Thanks!