ADAPTIVE K-MEANS METHOD FOR SEGMENTING IMAGES UNDER NATURAL ENVIRONMENT

This paper evaluates the performance of two conventional clustering-based segmentation methods and proposes an improved method for segmenting images captured under natural environment. Image segmentation refers to a process that separate area of interest from the background with the aim to extracts object of interest only for further image analysis. However, the segmentation process is very challenging for experiment conducted in outdoor environment due to the non-uniform illumination. Different illuminations produce different colour intensity for the object surface which leads to inaccurate segmented images. The widely used clustering-based segmentation methods are K-means and Fuzzy c-means (FCM). However, both methods have several limitations in producing good quality segmented images of objects that are exposed to the natural illumination. Therefore, this paper proposes an improved clustering-based segmentation method (Adaptive K-means) that is able to partition natural images accurately. The performance of three segmentation methods are evaluated on fruit images and compared quantitatively using similarity index (SI) and Tanimoto Coefficient (TC). The results show that Adaptive K-means has the ability to produce more accurate and perfect segmented images compared to the conventional K-means and FCM.