ABSTRACT
In classification, one is interested in the category of a sample. Classification problems arise in a myriad of areas—iris recognition, gait recognition, face recognition, speech recognition, character recognition, etc. Classification is used extensively in medicine—electroencephalography (EEG) signals are used to routinely identify stroke and Alzheimer’s, and sometimes, they are even used for predicting the subject’s thought in brain–computer interfaces. Sparse representation-based classification (SRC) assumes that the training samples of a particular class approximately form a linear basis for a new test sample belonging to the same class. The block sparse classification (BSC) approach sounds effective for general-purpose classification problems and was shown to perform well for simple classification problems. In BSC, all the training samples from the same class have the same class label. There are several non-linear extensions to the SRC approach. The linearity assumption in SRC can be generalized to include non-linear (polynomial) combinations.
