Filter Bank Techniques for Signal and Image Processing

Sammanfattning: This thesis consists of five parts, each of which considers methods based on filter banks for different applications in signal and image processing. In Part I, a non-parametric estimation method is proposed using filter banks to estimate signals with varying waveforms and arrival times. This method, which provides a linear, robust estimate of the signal, is based on a series expansion by a set of basis functions. Using a uniform and a non-uniform filter bank, the method is applied to estimation of an event-related signal. In Part II, a time-frequency distribution method for the analysis and synthesis of non-stationary signals is presented, incorporating a filter bank and the Wigner-Ville distribution (WVD). This method gives a representation of a signal in the time-frequency plane that suppresses the cross-terms and noise of the WVD, while retaining high energy concentration of the signal. Part III introduces an image enhancement method based on nonlinear multiscale reconstruction. By using a 2D-filter bank with a pyramidal scheme and different types of edge detectors, it is shown that the presented image enhancement method not only reduces noise, but also preserves the sharpness of edges. Part IV presents the design and application of a multirate digital filter bank. The impulse response of a system is estimated and multi-channel echo cancelling is performed employing multirate operations through decimation and interpolation. The presented approach is expected to yield high echo suppression and fast convergence for the filter weight adaptation by means of multirate parallel processing of the filter bank. In Part V, the detection of transient signals using filter banks and higher-order statistics (cumulants) is studied. It is shown that the use of the higher- order cumulants with a filter bank can be effective for the detection purpose.