Hairiness is a concept describing the amount of hairy fibers (hairs) protruding from a yarn core in different spatial orientations and shapes. Most image-based hairiness assessment methods measure hairs by projecting a yarn on a 2D image plane, which suffers from two major problems: 1) not detecting defocused hairs (fuzzy hairs) when hairs are out of the field of view of the imaging system and 2) miscalculating real lengths of spatially curved hairs. The objective of this research was to develop a new image-based hairiness measurement method to mitigate these problems. The proposed method included two tasks: yarn image segmentation and hairiness assessment. The first task was to improve the detection rate of fuzzy hairs with a hybrid algorithm combining double homomorphic filtering and region-growing algorithms. The second task was to establish a width-depth mapping model for defocused hairs to compensate measurable lengths of defocused hairs based on their width information. The experiment results demonstrated that the proposed segmentation algorithm can detect fuzzy hairs usually missed by the previously used algorithm, and can produce more accurate hair length measurements than the previous algorithm when compared to the corresponding manual measurements, which were considered as the gold standard in this study.