Robustness of noisy and blurry images segmentation
I. V. Gribkov, P. P. Koltsov, Н. В. Котович, A. A. Kravchenko, A. С. Куцаев, Andrey S. Osipov, A. V. Zakharov · Pattern Recognition and Image Analysis · 2009
In majority of applied pattern recognition systems, the first step of image preprocessing is segmentation. In the present work, we test robustness of four digital image segmentation methods with the aid of our PICASSO (PICture Algorithms Study Software, [1–3] (The report [3] was delivered in English at the PRIA-9-2008 international conference in Nizhni Novgorod (September 2008).) program system. According to the PICASSO’s general approach, the comparative study of quality of segmentation methods is fulfiled using special artificial test images. To evaluate the robustness, we calculate the difference between the result of segmentation of initial image and that of corrupted (noisy or blurry) image. The description of test images and testing procedures are given in the article. Our approach allows to clear up specific features and applicability limits of the segmentation methods under examination.