Fast fuzzy clustering of infrared images

Clustering is an important technique for unsupervised image segmentation. The use of fuzzy c-means clustering can provide more information and better partitions than traditional c-means. In image processing, the ability to reduce the precision of the input data and aggregate similar examples can lead to significant data reduction and correspondingly less execution time. This paper discusses brFCM (bit reduction by Fuzzy C-Means), a data reduction fuzzy c-means clustering algorithm. The algorithm is described and several key implementation issues are discussed. Performance speedup and correspondence to a typical FCM implementation are presented from a data set of 172 infrared images. Average speedups of 59 times that of traditional FCM were obtained using brFCM, while producing identical cluster output relative to FCM.