Consistent Sparse Representation based on Discriminative Dictionary Learning for Face Recognition

Feature engineering or feature extraction is an essential ingredient of machine learning recipes which can produce the huge difference to obtain an effective predictive model. This process is applied to transform an original input data into a high-level feature representation for training a predictive model. In image classification, sparse representation is one of the widely used techniques to represent images as linear combinations of an overcomplete dictionary. In this work, discriminability of samples from different classes and consistency of samples from the same categories are introduced by combining class labels into a dictionary learning process. However, it is time-consuming to compute the sparse coding as a result of sparsity constraint during optimization. Therefore, linear predictive model is applied to approximate the sparse representation features used to train a classification model. The performance of the proposed algorithm is compared to some existing discriminative sparse coding algorithms on benchmark dataset of face recognition. The results of the proposed method present improvement in both classification accuracy and computation time on the testing face images.