chang_report - Non-parametric Classification of Facial...

Info iconThis preview shows pages 1–3. Sign up to view the full content.

View Full Document Right Arrow Icon
Non-parametric Classification of Facial Features Hyun Sung Chang Department of Electrical Engineering and Computer Science Massachusetts Institute of Technology Problem statement In this project, I attempted to classify facial images based on various external characteristics, such as gender, expression, and accessories they are taking on. Rather than extracting any particular parameters describing faces, e.g., the distances among eyes, nose, and mouse, I used grey-scale face images themselves, fitted to 128x128 window, as the inputs. Dataset The dataset used for this project together with detailed description is available here at the course website. The dataset consists of 2,000 training face images (faceR, 1,997 of them labeled) and 2,000 test face images (faceS, 1,996 of them labeled). Because the image size is 128x128, each image can be considered as a data point in a huge dimensional space. The dimensionality reduction has been conducted using principal component analysis (PCA) on 100 sample faces, all from the training dataset, so each image can be represented by 99 eigenface coefficients, as well as the mean face. The composition of dataset is shown in Table 1. For example, notice that, in terms of expression, “funny” faces were significantly fewer than the other two classes and that few people wore glasses or bandana. One interesting thing is that no bandana image was included in the samples used to generate the eigenfaces.
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Table 1. Dataset composition gender expression male female serious smiling funny Eigenface generating data 61/100 39/100 Eigenface generating data 45/100 51/100 4/100 Training data (faceR) 1,150/1,997 847/1,997 Training data (faceR) 917/1,997 1,043/1,997 37/1,997 Testing data (faceS) 1,277/1,996 719/1,996 Testing data (faceS) 1,097/1,996 836/1,996 63/1,996 glasses bandana on off on off Eigenface generating data 4/100 96/100 Eigenface generating data 0/100 100/100 Training data (faceR) 59/1,997 1,938/1,997 Training data (faceR) 13/1,997 1,984/1,997 Testing data (faceS) 8/1,996 1,988/1,996 Testing data (faceS) 8/1,996 1,988/1,996
Background image of page 2
Image of page 3
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 12/04/2011 for the course ESD 1.124 taught by Professor Kevinamaratunga during the Fall '00 term at MIT.

Page1 / 9

chang_report - Non-parametric Classification of Facial...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online