Digital median filters

Several DSP algorithms need to remove high-frequency or impulsive noise while preserving edges, e.g., in speech and image processing applications: median filtering has been proved to be more effective for achieving this goal than other filtering techniques. Efficient architectural implementation for real-time applications involves a careful VLSI design, which takes into account modularity, regularity, adaptability, scalability, throughput, circuit complexity and fault tolerance. Four new architectural approaches are presented and evaluated in this paper to deal with different application and implementation constraints. They are: the serial-input polarizing median filter, the floating median filter, the pipelined polarizing median filter and the pipelined sorting median filter. The 1st and the 2nd architectures are based on majority voting, while the 3rd and the 4th ones are based on sorting techniques. All of them are designed so as to exhibit high scalability and to be easily pipelined for higher working frequencies.