Facial Keypoints Detection •

In this project, we are given a list of 96×96-pixel 8bit graylevel images with their corresponding (x,y) coordinates of 15 facial keypoints. We first adopt hold-out cross validation to randomly split the data set into a training set and a test set, so that we can develop our algorithm on the training set and assess its performance on the test set. Our algorithm first performs histogram stretching to enhance the image contrast by stretching the range of pixel intensity of each training image. Then, in order for noise reduction, we apply principal components analysis on the stretched images to obtain the eigenfaces. Using the resultant eigenfaces, we implement the mean patch searching algorithm with correlation scoring and mutual information scoring to predict the leftand right-eye centers for any query test facial images.