Size–density spectra and their application to image classification

We develop a density opening operator that is shown to satisfy the properties of an algebraic opening. This density opening enables the development of a number of variants of pattern spectra, which quantify the size or density information of a blob arrangement within the image. In contrast to regular morphological pattern size spectra, the proposed pattern spectra are spatially sensitive and robust to noise distortions. A pattern size–density spectrum or 2D signature was introduced and used for solving image classification tasks. Its application to the classification of real world medical images is illustrated.