IGO Subspace Learning is a new framework for visual learning particularly suitable for dealing with data corrupted by outliers such as occlusions and non-uniform illumination changes. IGO-methods could and should be used as an alternative to traditional subspace learning methods, such as L2-norm based PCA and LDA of pixel intensities, when robustness is of critical importance. They require the eigen-decomposition of simple covariance matrices and are as computationally efficient as their corresponding L2-norm intensity-based counterparts. In addition to that, our experiments have shown that IGO-methods outperform significantly popular methods such as Gabor features and Local Binary Patterns and achieve state-of-the-art performance for illumination- and occlusion-robust face recognition. For more details, please see
IEEE Transactions on Pattern Analysis and Machine Intelligence (accepted for publication).
The provided code performs subspace learning and feature extraction from a set of training images using Principal Component Analysis (IGO-PCA) and Linear Discriminant Analysis (IGO-LDA) of Image Gradient Orientations. It also provides methods for embedding new images in the subspace learned from IGO-PCA or IGO-LDA. A script file illustrating how to use the code for performing face recognition experiments is also provided. Should you make use of any of the code provided please cite the above paper.
IGOLearning Matlab Code: (requires matlab)