Jalba Andrei C, Wilkinson Michael H F, Roerdink Jos B T M
Institute for Mathematics and Computing Science, University of Groningen, 9700 AV Groningen, The Netherlands.
Microsc Res Tech. 2004 Sep;65(1-2):72-85. doi: 10.1002/jemt.20111.
A general framework for automatic segmentation of diatom images is presented. This segmentation is a critical first step in contour-based methods for automatic identification of diatoms by computerized image analysis. We review existing results, adapt popular segmentation methods to this difficult problem, and finally develop a method that substantially improves existing results. This method is based on the watershed segmentation from mathematical morphology, and belongs to the class of hybrid segmentation techniques. The novelty of the method is the use of connected operators for the computation and selection of markers, a critical ingredient in the watershed method to avoid over-segmentation. All methods considered were used to extract binary contours from a large database of diatom images, and the quality of the contours was evaluated both visually and based on identification performance.
提出了一种硅藻图像自动分割的通用框架。这种分割是基于轮廓的计算机图像分析自动识别硅藻方法中的关键第一步。我们回顾了现有成果，将流行的分割方法应用于这个难题，最后开发出一种能显著改进现有成果的方法。该方法基于数学形态学的分水岭分割，属于混合分割技术类别。该方法的新颖之处在于使用连通算子来计算和选择标记，这是分水岭方法中避免过分割的关键要素。所考虑的所有方法都用于从大量硅藻图像数据库中提取二值轮廓，并基于视觉和识别性能对轮廓质量进行评估。
