Linguistic operators of the Mathematical Morphology applied to the segmentation of images

AGUSTINA BOUCHET; JUAN IGNACIO PASTORE; MARCEL BRUN; VIRGINIA BALLARIN

Monterrey

Taller; VII Meeting of the Iberian-American Network on Multi-Criteria Decision Analysis (Red-M); 2014

CLAIO

Magnetic Resonance Imaging is a powerful tool for the early diagnosis of different neurologic diseases. This kind of images has high noise levels and imprecision in the definition of their structures. In this work we propose the use of the linguistic operators of the Compensatory Fuzzy Mathematical Morphology to improve the segmentation of lateral ventricles. We compare the results of these new implementation to the traditional Mathematical Morphology (MM) operators and to the clasical operators of the Compensatory Fuzzy Mathematical Morphology (CFMM). To compare their performance we implement a classic Mathematical Morphology segmentation algorithm and a Compensatory Fuzzy Mathematical Morphology version of the same segmentation algorithm. All of the methods show a high performance, with good spatial localization of the ventricles. Both methods based on CFMM has faster running time, but use all the information available in the gray level image. The method based on MM works on the binarized image, that may lead to information loss relying a good choice of the threshold. Once we designed the algorithm, we evaluate the results obtained from the use of each methodology. Their performance was studied on a 160 images set, besides gold-standard images, provided by the FLENI institute. The percentages of correctly segmented pixels for MM were 94.71%, 97.46% for clasical CFMM and 99.71% for linguistic operator of the CFMM.