Fuzzy image segmentation considering surface characteristics and feature set selection strategy

The image segmentation performance of any clustering algorithm is sensitive to the features used and the types of object in an image, both of which compromise the overall generality of the algorithm. This paper proposes a novel fuzzy image segmentation considering surface characteristics and feature set selection strategy (FISFS) algorithm which addresses these issues. Features that are exploited when the initially segmented results from a clustering algorithm are subsequently merged include connectedness, object surface characteristics and the arbitrariness of the fuzzy c-means (FCM) algorithm for pixel location. A perceptual threshold is also integrated within the region merging strategy. Qualitative and quantitative results are presented, together with a full time-complexity analysis, to confirm the superior performance of FISFS compared with FCM, possibilistic c-means (PCM), and suppressed FCM (SFCM) clustering algorithms, for a wide range of disparate images