Reddy, C., F-I. Liu, J. Zurdo, and F. B. Dazzo
Presented at the All Scientist Meeting (2002-10-04 )
Digital color imaging for quantitative image analysis is a potentially powerful methodology for studies of in situ microbial ecology, but the technology needed to fully utilize this approach is at a rudimentary stage. The color pixels that comprise microbes in RGB images can be used to differentiate their Gram’s reaction, (live/dead) membrane integrity, metabolic activity, phylogeny, and autecological biogeography. Spatial position, morphological diversity and abundance of microbes are other important in situ ecological features that can potentially be extracted from color images of microbial communities. However, one faces several major problems in editing digital color images of microbes before image analysis, including the presence of complex, noisy backgrounds, pixels of foreground objects of interest that vary significantly in their RGB values, microbes containing internal dark pixels with RGB values included in background, and bright halos of background pixels surrounding cell contours. Because various combinations of these problems may exist in images containing a high density of microbes, much time must be spent manually editing them before one can start to extract useful information. Various image processing routines are available that provide a simple color threshold based on sampling only a single color pixel. But this approach most frequently fails to segment most foreground microbes of interest because of the problems stated above. Here we introduce a new CMEIAS Color Recognition program that addresses these major image editing problems in order to segment the colored microbes of interest. Our approach uses color recognition, i.e., classification of pixels based on the color information from the digitized color image, to identify different selected regions representing the microbes of interest. The novelty of this new system is the interactive environment of the sampling module, where the user can zoom in and select multiple target pixels from many cells in order to segment the foreground regions of interest in a confusing background. The system applies the RGB information from the selected set of multiple color pixels to a region-growing algorithm that finds the boundary of the foreground objects. Additionally, various routine image processing and computer vision algorithms, e.g., image smoothing, sharpening, closing, edge detection, color dilation and erosion, are included to enhance the output of the edited color image. Bacteria colored by different molecular probes can also be segmented individually in the same microbial community image. We propose that color segmentation achieved by this new CMEIAS program provides an efficient first stage of image editing, and anticipate that the operating principal of this tool will open many new opportunities for quantitative image analysis of microbes in digital color images. However, as with all applications of digital image analysis, color images of the microbial community must be of high quality as a prerequisite. Our first application of this CMEIAS Color Recognition program, illustrated in our other CMEIAS poster for this meeting, was to produce segmented color images for analysis of the spatial distribution of color-coded bacteria that provide a source of quorum sensing cell communication molecules in situ during their colonization of plant roots
Back to meeting | Show |