Efficient segmentation of degraded images by a neuro-fuzzy classifier

Abstract

The segmentation of degraded images has always been a difficult problem to solve. Efficient object extraction from noisy images can be achieved by neuro-fuzzy clustering algorithms where noise pixels are identified during the clustering process and assigned low weights to avoid their degradation effect on prototype validity. We present a new approach to noise reduction prior to segmentation by using a two-step process named the AFLC-median process. This new two-step process has been specifically tailored to remove speckle noise. The first step is to use an AFLC (adaptive fuzzy leader clustering) network that has been designed to follow leader clustering using a hybrid neuro-fuzzy model developed by integrating a modified ART-1 model with fuzzy c-means (FCM). This integration provides a powerful, yet fast method for recognizing embedded data structure. In speckled imagery, AFLC is used to isolate the speckle noise pixels by segmenting the image into several clusters controlled by a vigilance parameter. Once the speckles have been identified, a median filter is used, centered on each speckle noise pixel. The resulting image, after undergoing the AFLC-median process, demonstrates a reduction in speckle noise whilst retaining sharp edges for improved segmentation.