Deconvolution regularized using fuzzy c-means algorithm for biomedical image deblurring and segmentation
Benoît Lelandais, Frederic Ducongè · 2015
We address deconvolution and segmentation of blurry images. We propose to use Fuzzy C-Means (FCM) for regularizing Maximum Likelihood Expectation Maximization deconvolution approach. Regularization is performed by focusing the intensity of voxels around cluster centroids during deconvolution process. It is used to deconvolve extremely blurry images. It allows us retrieving sharp edges without impacting small structures. Thanks to FCM, by specifying the desired number of clusters, heterogeneities are taken into account and segmentation can be performed. Our method is evaluated on both simulated and Fluorescence Diffuse Optical Tomography biomedical blurry images. Results show our method is well designed for segmenting extremely blurry images, and outperforms the Total Variation regularization approach. Moreover, we demonstrate it is well suited for image quantification.