Face-Detection

Face detection (class and psoe) using various Classifiers. Dimentionality Reduction using PCA and MDA

Classifiers implemented from scratch

• Bayes
• kNN
• SVM (Linear, RBF, Polynomial kernels)
• AdaBoost with SVM

Author

Siddharth Telang (stelang@umd.edu)

Subject Code

CMSC828C/ENEE633 Project 1

Programming language used: Python3+

Dependencies (to be installed through pip):

1) sklearn(used only for PCA): pip install sklearn
2) matplotlib: pip install matplotlib
3) numpy: pip install numpy
4) scipy (.mat to python): pip install scipy
5) cvxopt (quadratic solver): pip install cvxopt

Contents:

1) Code files:
- Helper functions (my_pca, my_mda, my_lda, helper_functions, svm_helper) used by main files
- bayes_classifier_subject, knn_subject: Subject label classification
- bayes_classifier: bayes' classifier implementation
- classify_pose_Bayes, classify_pose_KNN: Pose identification for Data set 1
- svm_classifier, adaboost: Pose identification for Data set 1
2) Report
3) Figures - all plots
4) Data folder containing the dataset

Steps to run the code:

• Please ensure this to be the current working directory.
• Various commands with different permutations are mentioned below.
• You may use this on the command prompt and terminal.
• A choice of choosing among pca or mda is provided. Feel free to update if required, only one of them can be set to True at a time.
• trainingSize parameter can be altered to test on various training and testing size
• kernel parameter is provided to select between 'rbf', 'poly', and 'linear' kerel svm
• iterations parameter can be updated for more number of iterations in boosted svm

1) Subject Label classification

• Bayes Classifier

python bayes_classifier_subject.py -dataset Data/data.mat -subjects 200 -types 3 -pca True
python bayes_classifier_subject.py -dataset Data/pose.mat -subjects 68 -types 13 -mda True
python bayes_classifier_subject.py -dataset Data/illumination.mat -subjects 68 -types 21 -mda True

• k-NN

python knn_subject.py -dataset Data/data.mat -subjects 200 -types 3 -pca True
python knn_subject.py -dataset Data/pose.mat -subjects 68 -types 13 -mda True
python knn_subject.py -dataset Data/illumination.mat -subjects 68 -types 21 -mda True

2) Neutral vs Expression identification

• Bayes Classifier

python classify_pose_Bayes.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 200 -pca True
python classify_pose_Bayes.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 200 -mda True
python classify_pose_Bayes.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 300 -mda True

• k-NN

python classify_pose_KNN.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 200 -pca True
python classify_pose_KNN.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 200 -mda True
python classify_pose_KNN.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 100 -mda True

• Kernel SVM

python svm_classifier.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 300 -pca True -kernel rbf 
python svm_classifier.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 300 -mda True -kernel rbf 
python svm_classifier.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 300 -pca True -kernel poly
python svm_classifier.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 300 -pca True -kernel linear

• Ada-Boost (Linear SVM)

python adaboost.py -dataset Data/data.mat -subjects 200 -types 2 -trainingSize 300 -pca True -kernel linear - iterations 10