This article is a continuation of digitization of images from the microscope into the video camera. Why should microscopists (or any other scientist) care about or use computer-based image processing and measurement? There are several reasons; a few are: to assist the human viewer in observing or communicating information in images; to minimize human bias based on wish or expectation; to introduce rigor into the process of obtaining quantitative information as a substitute for anecdote; and not least, to make us better and more aware viewers of images. Unassisted human vision is rarely a reliable scientific tool. Henry David Thoreau said “The question is not what you look at, but what you see.”
John Russ has taught hands-on courses and extended workshops in image processing and analysis to more than 3000 students, worldwide, over the course of his career. His one-day tutorials and lectures, sponsored by various professional societies and other organizations, have reached several thousand more. But the need to have a basic understanding of these topics is far wider than he can ever reach in person. Potentially everyone working with images, and certainly that includes every microscopist, needs to be aware of the possibilities (and limitations) of computer-based image processing and measurement. The descriptive reviews and interactive tutorials in this section cover most of the topics that the author discusses in typical one-day tutorials.
Overview of Image Processing and Analysis - Image Processing operates on images and results in images, with changes intended to improve the visibility of features, or to make the images better for printing or transmission, or to facilitate subsequent analysis. Image Analysis is the process of obtaining numerical data from images. This is usually accomplished by a combination of measurement and processing operations. The data may subsequently be analyzed statistically, or used to generate graphs or other visualizations.
Correcting Image Defects - Most images include some imperfections, the result of the inherent limitations in illumination, optics, camera or the specimen itself. Many of these can be improved by processing. If color images are being acquired, it is first of all important to understand something about color representations. Then the removal of noise may be required, using methods that depend on the noise source. After correction for nonuniform illumination, the contrast and brightness can be adjusted for optimum visibility of detail. Finally, limitations in image focus may be addressed.
Enhancement of Image Detail - The procedures described and illustrated above are all intended to compensate for various limitations and defects that arise in acquiring a digitized image. Their goal is to produce a correct representation of the original scene. By proper use of many of the same tools, it is also possible to enhance the visibility of some details and information in the image. This is accomplished by removing or suppressing other information (which is not currently of interest), so that what remains is more readily seen by human vision, and/or more readily isolated for measurement.
Binary Images - Thresholding an image converts a gray scale or color original to a black-and-white version that distinguishes feature(s) from background. The features are composed of those pixels that are of current interest for some kind of measurement procedure, whereas the background consists of the pixels that are not of current interest. Morphological and Boolean operations are applied to binary images to selectively and accurately delineate the features, which are usually assumed to correspond to some objects or structure that is present in the scene. Of course, the original image should always be kept because in the future the objects or structures of interest may change!
Measurements - Measurements can be classed generally into two groups: those for the entire image or scene (usually based on stereological procedures and extrapolated to the entire specimen represented by the sample), and those performed on each individual feature or object present (measures of size, shape, position and color or density that are usually summarized statistically or used for feature recognition). Read more