Combining reconstructive and discriminative subspace methods for robust classification and regression by subsampling
Linear subspace methods that provide sufficient reconstruction of the data, such as PCA, offer an efficient way of dealing with missing pixels, outliers, and occlusions that oft...