Pre-processing of images in an automated chromosome analysis system

摘要

Pre-processing of human chromosome images in a computer-based automatic classifier is assessed, comparing five different types of filtering, and no filtering. Assessment is by visual appearance of the images after pre-processing, by measurement of the error in an automatic centromere finder in the classifier and by the error in initial segmentation of object images. Smoothing parallel to the image contours followed by a Laplace filter produces the best results with regard to centromere finding, but at the expense of fragmenting a greater proportion of chromosome images than the other techniques.