Title :
Color image segmentation based on Markov random field clustering for histological image analysis
Author :
Meas-Yedid, Vannary ; Tilie, Sorin ; Olivo-Marin, Jean-Christophe
Author_Institution :
Quantitative Image Anal. Unit, Inst. Pasteur, Paris, France
Abstract :
In order to characterise the virulence factors of different Mycobacterium tuberculosis strains responsible for the tuberculosis disease, the quantification, by cell counting, of immune cell recruitment is necessary. However this task by microscopic observations is very tedious and difficult to reproduce. Hence we propose an automatic counting approach, consisting in color image segmentation to discriminate three regions: cell nuclei, immune cells and background, followed by the extraction of each cell entity. For color segmentation, a Markov random field clustering approach taking simultaneously into account both color and spatial information is chosen. Our technique was successfully applied to several color images of different strains, and an evaluation of the results has been performed, showing the robustness of the method against noise, marker color changes, illumination changes and blurring.
Keywords :
Markov processes; biological tissues; cellular biophysics; diseases; image colour analysis; image segmentation; medical image processing; microorganisms; patient diagnosis; Markov random field clustering; Mycobacterium tuberculosis strains; automatic counting approach; blurring; cell counting; cell nuclei; color image segmentation; histological image analysis; illumination changes; immune cell recruitment; immune cells; marker color changes; noise; spatial information; tuberculosis disease; virulence factors; Capacitive sensors; Colored noise; Data mining; Diseases; Image color analysis; Image segmentation; Markov random fields; Microscopy; Performance evaluation; Recruitment;
Conference_Titel :
Pattern Recognition, 2002. Proceedings. 16th International Conference on
Print_ISBN :
0-7695-1695-X
DOI :
10.1109/ICPR.2002.1044879