Face detection by adaboost cascade

Face detection is a challenging task as different people have features due to their race and other important factors to consider are the variation in the image plane as well as the lighting variation. Classifiers in a cascaded structure are able to increase the detection performance of a single, weak classifier. This project focuses on training a classifier that would be part of a two classifier cascaded structure to correctly identify face and non-face images using Matlab as the programming language. The classifier was trained using the appearance based method by using many different face and non-face images and calculation was done based on the concept of Asymmetric Principal Component Analysis (APCA) from Professor Jiang’s paper [1]. Two methods were used to train and test the classifier with the training and testing datasets and the performance of the classifier was evaluated and analyzed under different parameters to determine the condition for optimal performance. The results obtained from testing the cascade classifier on actual digital images were of acceptable level as the images identified as “face” had a true positive rate of 99.66% and a false positive rate of 0.34% when the threshold value, ao and m value were set for a low false positive rate. The process of doing this project has led to an understanding of how a classifier is trained and tested.Bachelor of Engineerin