Restrained neighborhood analog median filters

The implementation of highly parallel median filters for image processing using a reduced interconnection complexity is presented. The performance of the noise removal process is analyzed, and the results are compared with the previously reported. Two analog median filters were used with this technique: a median filter that is based in a transconductance comparator, and a subthreshold operation median filter. Simulation results of a 78 /spl times/ 89 image corrupted, with 15% salt and pepper noise, are shown. All the simulations were made using BSIM3 level 49 model and 1.2 /spl mu/m MOSIS parameters.