A coupled-contour model for effective segmentation of cotton cross-sections in wide-field microscopic images

Abstract

Image analysis of a fiber cross-section can provide direct measurements for cotton maturity. Effective segmentation of fiber contours in a cross-sectional image is paramount for accurate fiber geometrical measurements. In a wide-field microscopic image, the adhesion, breakage, and ambiguity (low contrast or noise) of fiber contours make the segmentation rather challenging. This paper presents a new approach for contour segmentation that takes advantage of the shape features of the triple concentric contours, called the coupled-contour model (CCM), of a cross-section, and a CCM-based algorithm developed to locate, split, merge, and refine fiber contours based on the established rules concerning contour features. For a wide-field microscopic image (12 megapixels), this CCM-based algorithm could detect >500 fiber cross-sections with a recall rate of 93.53% and a precision rate of 98.13%, and reduced the errors in maturity measurements by 50%.