A dynamic threshold algorithm using connectivity analysis

In order to recognize an object which is embedded in a complex scene, it first must be located and its contour well defined. Such a preprocess, one of Digital Image Processing techniques, is called segmentation.

In this thesis, some methods to perform binary segmentation, by dynamic thresholding the digitized images, were developed. Results from these methods are presented; two of them were used in a character extraction application, based on contour analysis. The darkness relation between object and background plays an important role in the segmentation process. There is no general algorithm which will work for all images; so, every known characteristic from a given set of pictures should be included in a particular implementation of the appropriate method. Hints in selecting a method are also given.