New Method Of Image Analysis Research Articles

Classification of electron microscopical images of randomly oriented bio macromolecules

New methods of image analysis in which aligned images of bio-macromolecules are submitted to multivariate statistical analysis in the form of correspondence analysis (CA) have largely facilitated the understanding of mixed populations of images. CA determines the main directions of interimage variance and gives the images new coordinates along the axes thus determined. A large reduction in the amount of data is obtained: instead of, for example, 64x64=4096 density values per image, each image is now characterized by eight factorial-axis coordinates at the most! The problem of classification of the images has therefore come closer; however, the essential final question still must be answered: " How does one objectively take images together in groups such that the classes of the 'partition' obtained represent the 'best' possible classification?" In the case of randomly oriented particles we would like the classes to contain only molecules in one particular orientation relative to the support film ("projection classes").