Contextual possibilistic modeling of pixellic knowledge for tumor segmentation in mammographic images

In this paper a novel possibilistic knowledge modeling at the pixel level, is proposed. This model consists on the use of the spatial contextual information at the level of each pixel, in order to evaluate a local based possibility distribution, resuming the pixel information. The proposed possibilistic modeling approach performance is evaluated through a pixel classification of both synthetic image and 5 mammographic images. Its performance is compared with three relevant reference methods: classic Bayesian approach and Markov Random fields approach with two optimization technics: Iterated Conditional Modes (ICM) and simulated annealing (RS). Our approach outperforms the other methods, in terms of recognition rate, by 94.84%, against, respectively, 93.88%, 85.10% and 84.67%. In addition, the proposed possibilistic modeling approach showed an interesting behavior of stability and allowed a better visually classification quality compared to other methods.