Abstract

In vidco scquence processing,shadow remains a major source of error lor object segmenlalion.Traditional methods of shadow removal are mainly based on colour difference thresholding belween the background and currenl images.The applicalion of colour fillers on MPEG or MJPEG images,howcver,is oftcn crroncous as thc chrominancc information is significantly rcduccd duc to comprcssion.In addition,as thc colour attributes of shadows and objccts arc oficn very similar,discrctc thrcsholding cannot always providc rcliablc rcsults.This papcr prescnts a novcl approach for adaptivc shadow rcmoval by incorporating four different filters in a neuro-fuzzy framcwork.The neuro-fuz.zy classifier has the ability of real-time self-adaptation and training, and its performance has been quantitatively assessed with both indoor and outdoor vidco scqucnccs.