Automatic multi-thresholdable image segmentation by using separating bipoints

We present a method of segmentation of multi-thresholdable images. Many algorithms try to use information provided by the contrast of the image. These techniques give different kinds of results depending on images. From this idea we propose a different way to use the contrast of the image. It consists of looking for bipoints corresponding to the normals to the most striking boundaries. The thresholds take their values within the intervals defined by these bipoints. To obtain the best thresholds, we look for separators which cut at best the intervals. This method is an improvement of methods based on the histogram or on co-occurrence matrix but with a different approach. It provides a better detection of small regions with high contrast, difficult to detect with previous methods. This method has several advantages. First it is basically simple and more intuitive than other mathematical approaches. In addition it allows fuzzy segmentation in case of noisy boundaries and it gives good results with a set of several test images.