A Pretopological Approach for Image Segmentation and Edge Detection

We present an approach based on a pretopological formalism that allows the mathematical modeling of image segmentation by region growing. The choice of pretopology is motivated by the fact that it has less axioms than the topology which facilitates its adaptation to discrete spaces and in particularly image processing. In our approach, the pretopological adherency function associated to the pretopological structure is defined by a criterion of homogeneity. We apply our approach to the extraction of handwritten information on check background with images of scene and to edge detection. The aggregation from chosen initial germ ends to a closed part of the image composed by a stroke (a line of handwriting) or a line of contours. The evaluation, undertaken on 60 checks with various background images and different images, gives 92% of good results for the extracted handwriting with a complete elimination of the background and good results for edge detection.