Least-squares design of 2-D sparse nonseparable filter banks using transformation of variables: A greedy approach

We present in this paper a design method for two-dimensional two-channel sparse nonseparable FIR filter banks using transformation of variables. In a first phase the transformation function is designed in a least-squares fashion. The second stage of the algorithm introduces the sparseness in a greedy manner by removing at each step the minimum coefficient of the filter in absolute value. The obtained results show that the sparse filter banks offer the possibility of choosing the complexity of the filters and have also better stopband energy.