Local entropy edge detection in digital images
David Charles Martin · Proceedings annual meeting Electron Microscopy Society of America · 1988
Detection of edges in digital images is an important task for feature recognition and interpretation. Typically, edge detection schemes involve difference operators such as the gradient or Laplacian which depend on the rate of change of brightness in an image. Recently, Shiozaki has shown that a local entropy operator can be useful for edge detection in digital images. This method is simple, rapid, and the resulting image can be interpreted as a measure of the local “information” content of the original data. Here, we investigate the applicability of this approach to digital electron micrographs. The definition of the entropy for a probability distribution pi is S=-Σ pi in Pi. For an image, the conditional probabilities pi are defined as the fraction of total flux which is in a given pixel. If the intensity of the image at a pixel i is fi, then pi=fi/Σfi. The entropy has a maximum when all the Pi's are equal, corresponding to the case of least configurational information.