Section: New Results
Image Segmentation, Registration and Analysis
Wavelet-based non-local means denoising
We have proposed here a fully automatic 3D blockwise version of the Non Local (NL) Means filter with wavelet sub-bands mixing. The proposed wavelet sub-bands mixing is based on a multi-resolution approach for improving the quality of image denoising filter. Quantitative validation was carried out on synthetic datasets generated with the BrainWeb simulator. The results showed that our NL-means filter with wavelet sub-bands mixing outperforms the classical implementation of the NL-means filter in terms of denoising quality and computation time. Comparison with well-established methods, such as non linear diffusion filter and total variation minimization, showed that the proposed NL-means filter produces better denoising results.
Mapping asymmetries of bilateral objects represented by point clouds
We developed a method to automatically quantify the local asymmetries of bilateral structures represented by point clouds. Such data can be obtained for instance by laser scanning a human face. The method relies on the robust computation of the approximate symmetry plane of the object under study. Departure from perfect symmetry is then computed for each point of the cloud, allowing to map the asymmetries of the object under study. Spatial normalisation techniques are currently developed to allow for comparison of different populations of subjects (males/females, controls/schizophrenics, etc.).
Deformable Segmentation of vector and tensor field MRI using Graph-Cut
During the first half of year, I have made a review about the different existing segmentation methods in biomedical imaging. This review, available as an INRIA research report (number 6306), classifies the methods in five main fields which are contour-based, region-based, shape-based, graph-based and eventually structural-based approaches. For each of these approaches, we then explain the principal methods and illustrate them by giving few examples of each. The second half of the year was dedicated to the implementation of graph-cut based segmentation methods. Once done, we have looked at using the graph-cut paradigm for multimodal brain MRI and tensor-based images.