Vectorization of human pelvis objects in X-ray images

In medical diagnostics visual evaluation of an object or its image is necessary but time consuming operation. Well-known computer vision algorithms or their compilation, or even some new methods should be the right tool in increasing the speed and reliability of this process. This paper introduces situation in this domain and some experiments and their results in extraction of biomechanical parameters of human pelvis from x-ray images using combination of Hough transform for a line, for a circle (arc) and Canny edge detector. The main idea of an algorithm, which was created during this experiment, is to use different levels of noise filter thus making a balance between leaving too much noise and removing too much actual data. The basic steps would be: filter out most of noise and noisy objects using high filter's threshold value; find sharp and clear objects; narrow the set of possible parameters of noisy objects; apply noise filter with lower threshold value to the original image; find noisy objects. Experiment shows that algorithm works but it needs to be tested on reliability and some bindings with actual biomechanical parameters should be done (see [1-6]).