Block processing for rank order filtering using the rank order state machine architecture

Rank order filters are nonlinear filters which choose an output based on its rank within a one-dimensional window of sample inputs determined by sorting the inputs. These filters are constructed using running order sorter architectures which utilize the overlap between consecutive outputs to maintain a sorted list of the inputs within the sample window. Additional speedup can be achieved in the rank order filter structures using block processing techniques. Block processing transforms a single-input single-output structure into a parallel-input parallel-output structure. It is difficult to apply block processing to most running order sorters because of the inherent feedback in their structures. A new class two architecture for rank order filtering, the rank order state machine, has been introduced. The authors discuss how block processing can be effectively applied to the rank order state machine architecture. They also extend this architecture to support block structures for stack filtering. Several architectures for rank order filtering and stack filtering using the rank order state machine architecture are presented.<>