Microscopy Cell Image Segmentation Based On The Robust Level Set Method

As the basis of cell counting,cell tracking and cell morphology analysis,cell automated segmentation is widely used in medical,biological and environmental monitoring applications.Up to now,various cell image segmentation algorithms have been progressively proposed and successfully applied to different cell images.However,due to the difference in cell morphology and characteristic,ones have to design different algorithms for different scenarios.Due to internal genetic material,bacillus cells microscopy images are inhomogeneous,and in addition,the cells are densely distributed and even cause adhesion.The level set method,i.e.,geometric active contour model,represents planar curve by using the zero level set of level set function defined in a high dimensional space,and could flexibly handle the topological changing of curves.It designs an energy functional of the level set function,converts the segmentation into an optimization problem of the energy functional,and finally realizes image segmentation.Since it could track smoother contour of targets,and possesses flexible capability of handing topological change of curves,level set methods are widely concerned and studied in the field of cell image segmentation.Level set methods drive the curve evolving by using region and edge information of cells to be segmented,and have potentiality to deal with the segmentation of the cells with blur edges and inhomogeneity.Study of the proportion of bacillus cells in organism is of great significance for the control of conditional pathogens.However,limited by sampling conditions,the cells in the division cycle are usually densely distributed,and the noise pollution is serious,which causes the segment result is not satisfied,and it is easy to lead to under-segmentation or even failure of segmentation.Aiming at the above problems and considering the characteristics of the microscopy Bacillus cell dataset,we re-design the stop function of the level set function as well as the energy functional.The main work is summarized as follows.(1)A level set method based on multi-scale edges is proposed.Aiming at the undersegmentation of level set methods on images with adhesive cells,we designed a level set method based on multi-scale edge stop function.Firstly,the LBP feature of the original image is extracted to locate seed points of the cells,based on which the initial curve is placed.Secondly,a multi-scale edge stop function is designed to describe the cell edge in more details.Finally,the curve evolves in an expanded manner and stops at the edge of the cell,driven by a stop function based on multi-scale edges.The method realizes the automatic initialization of curve by means of the seed points,and the multi-scale edge stop function is used to improve the driving force of cell edge on evolution curve,which can effectively avoid the under-segmentation with adhesive cells.(2)A noise robust multi-task level set method is proposed.To well handle the segmentation of cell images with noise interference and inspired by the total variation de-noising model,we designed a new energy functional.By optimizing the proposed energy functional,the denoising and segmentation of cell images could be realized simultaneously.The method combines image de-noising and segmentation together by the energy functional.The denoised image is used to segment the task,and the segmentation result is inversely applied to image denoising.In this way,cell segmentation can be achieved while image denoised,simultaneously,segmentation also affects de-noising.The method effectively improves the robustness against the noise interference at the same time,makes the de-noising result more suitable for segmentation.In summary,this paper improves current level set methods by designing a new curve evolution stop function and energy functional,provides cell segmentation with new ideas and methods,and contributes to the development and application of level set methods.