| abstract |
A method and apparatus for classifying an image is proposed. The image is first preprocessed by for example adjusting its orientation, selecting an area of interest and removing artifacts, and normalizing the brightness. Subsequently the image is processed by applying a wavelet decomposition and values for a scalar feature of the wavelet decomposition are measured. This value is binned and probabilities for different classifications of the image are determined based on the number of images from a training set that fall into the same bin that have each classification. Improved classifiers are constructed by aggregating the results for different features, and a network of improved classifiers, in which images are processed by later classifiers according to the results from earlier classifiers, is used. |