Segmentation of overlapping cells in cervical smears based on spatial relationship and Overlapping Translucency Light Transmission Model. (December 2016)

• Record Type: Journal Article
• Authors: Zhang, Jianwei; Hu, Zhenpeng; Han, Guoqiang; He, Xiaozhen
• Is Part Of: Pattern recognition. Volume 60(2016:Dec.)
• Page Start: 286
• Page End: 295
• Publication Date: 2016-12
• Subjects: Cervical overlapping cell -- Overlapping Translucency Light Transmission Model -- Spatial relationship -- Cell overlapping matrix -- Beer–Lambert law; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2016.04.021
• Languages: English
• ISSNs: 0031-3203

Abstract

Abstract: Overlapping cell segmentation in cervical smear images is a difficult task due to the shape multiformity and color proximity of the cells. In this paper, we propose a segmentation approach by using the spatial relationship of the non-overlapping and overlapping areas as well as Overlapping Translucency Light Transmission Model (OTLTM) to segment the overlapping cells in these images. The spatial relationship, which denotes the overlapping area locates in the middle ground of the non-overlapping areas, reflects the overlapping area can be accurately gained by the precise non-overlapping areas. After removing the background by threshold technique, a fragmentation method by using mean shift and watershed is adopted to divide the overlapping cells into fragments according to the similarity of their colors. The fragments belong to a single-tier individual cell, or to the overlapping area between two cells. We firstly construct the initial fragment collections of non-overlapping areas based on the Voronoi diagram, then the initial collections are optimized by using the initial cell overlapping matrix based on the spatial relationship, and OTLTM based on Beer–Lambert law, which states the relationship between the transmittance, attenuation coefficient of a kind of material and the distance the light travels through it. The cell overlapping matrix is accurately reconstructed by the optimized set of the non-overlapping areas. We obtained the segmentation result by combining the cell overlapping matrix and the optimized set of the non-overlapping areas. The experimental results show that the proposed method can give an impressive performance. Besides cervical smear images, these proposed techniques can be utilized in segmenting translucent objects from other kinds of images. Highlights: We obtain the fragments by coarse segmentation and base these fragments to achieve the result. We propose Overlapping Translucency Light Transmission Model. We derive overlap color measurement formula, and provide overlapping fragment judge's theory. We propose spatial relationship of the non-overlapping and overlapping areas. Our method can also be applied in segmentation of translucent object from other images.