Modified K-nearest neighbour filters for simple implementation
David Z. Gevorkian, Karen Egiazarian, Jaakko T. Astola · 2002
The K-nearest neighbor (K-NN) filter introduced by Davis and Rosenfeld (1978) has been long time successfully used in application to noise smoothing problems. The main drawback of this filter is its very high computational time. In this paper we introduce a slightly modified version of the K-NN filter and show that these two filters have practically the same performance in the noise removal sense while our modification is simpler in implementation. A binary-tree search technique for the implementation off the modified K-NN filter is presented, and an efficient bit-serial architecture for implementation of this filter is proposed.