Algorithms and architectures for a class of non-linear hybrid filters

Abstract

Non-linear filters based on ranked order statistics form an important class of filters for image enhancement. They can smooth noisy images while retaining edge structures. Recently, a new class of non-linear filters called FIR-median hybrid (FMH) filters were proposed. They are more computationally efficient than the conventional median filter as the size of the window increases, since the number of ordering operations is independent of the size of window. Also, they can preserve fine details such as lines. In real time applications such as industrial inspection, techniques for fast computation of image processing filters, especially using parallel architectures, are of great interest. Recently there have been several inspection systems using pipeline architectures and various new algorithms have been proposed. In this paper, we describe algorithms for computing various classes of FMH filters using a pipeline image processing system equipped with hardware comparators, which are available on most commercial pipeline machines. These include the unidirectional, the bidirectional, and the multilevel FMH filters. The correctness of the algorithms is proven. The time and space complexities of the algorithms are analyzed.