HIGH RESOLUTION PURSUIT FOR FEATURE EXTRACTION

Abstract Recently, adaptive approximation techniques have become popular for obtaining parsimonious representations of large classes of signals. These methods include method of frames, matching pursuit, and, most recently, basis pursuit. In this work, high resolution pursuit (HRP) is developed as an alternative to existing function approximation techniques. Existing techniques do not always efficiently yield representations which are sparse and physically interpretable. HRP is an enhanced version of the matching pursuit algorithm and overcomes the shortcomings of the traditional matching pursuit algorithm by emphasizing local fit over global fit at each stage. Further, the HRP algorithm has the same order of complexity as matching pursuit. In this paper, the HRP algorithm is developed and demonstrated on 1D functions. Convergence properties of HRP are also examined. HRP is also suitable for extracting features which may then be used in recognition.