Semi-Automated Segmentation of Microbes in Color Images

摘要

The goal of this work is to develop a system that can semi-automate the detection of multicolored foreground objects in digitized color images that contain complex and very noisy backgrounds. Although color image segmentation is considered a general problem, our application is microbiology where various colored stains are used to reveal information about the microbes without cultivation. Instead of providing a simple threshold, the proposed system offers an interactive environment whereby the user chooses multiple sample points to define the range of color pixels comprising the foreground microbes of interest. The system then uses the color and spatial distances of these target points to segment the microbes from the confusing background of pixels whose RGB values lie outside the newly defined range and finds the boundary of the foreground microbes using region-growing and mathematical morphology. Some other image processing methods are also applied to enhance the resultant image containing the colored microbes against a noise-free background. The prototype performs with 98% accuracy on a test set compared to manually edited ground truth data. The system described here will have many applications in image processing and analysis where one needs to segment typical pixel regions of similar but non-identical colors.