[home] [Personal Program] [Help]

---

Session: Poster session II
Session starts: Thursday 24 January, 16:00

---

Jyotirmoy Banerjee (Erasmus MC, Rotterdam. The Netherlands)
Adriaan Moelker (Erasmus MC, Rotterdam, The Netherlands)
Wiro Niessen (Erasmus MC, Rotterdam, The Netherlands)
Theo van Walsum (Erasmus MC, Rotterdam, The Netherlands)

Abstract:
Visual tasks such as detection, localization, categorization, and recognition are important subjects of study in medical image analysis. These tasks are often difficult due to apparent within-class inhomogeneity. Invariant image descriptors extract information from images which is invariant to the variability introduced due to the image formation process. One class of such descriptors is texture patterns. Texture has received considerable attention [1] with application in areas of medical imaging [2]. The local binary patterns [LBP], introduced by Ojala et al. [3], are an efficient method for texture description in 2D. One of the important aspects of this texture descriptor is its rotational invariance. The aim of our work is to extend the conventional LBP and its rotational invariant property mentioned in [3], to a 3D paradigm. Our method is capable of rotationally invariant description of landmarks or regions in 3D using LBP. The LBP in 3D requires a spherical sampling, which is represented in a spherical harmonics framework [4]. The framework helps in obtaining rotation invariant representation. Further, the region information is collected to a set of histograms that are invariant to rotation. The similarity between any two regions can be computed using the Chi-square distance measure [5] between the corresponding set of histograms. Concluding, we presented a method for rotationally invariant 3D LBP, using spherical harmonic decomposition. We applied the method on vessel-like phantom data and a clinical dataset, with encouraging results. More in-depth analysis and integration of complementary measures is part of future work.