Bernd Hamann - Publications Authored, Co-authored, or Co-edited
(Including Accepted Publications to Appear)
[504] Linares, O.A.C., Belizario, I.V., Batah, S.S., Hamann, B., Fabro, A.T., Azevedo-Marques, P.M. and Traina, A.J.M. (2024), RadPleura: A radiomics-based framework for lung pleura classification in histology images from interstitial lung diseases (pdf), in: Golemati, S. and Konofagou, E.E., eds., Proceedings of 21st IEEE International Symposium on Biomedical Imaging (ISBI 2024), IEEE Xplore Digital Library, IEEE Press, Piscataway, New Jersey.
[505] Zhang, X., Hamann, B., Wang, D., Zhang, H., Sheng, J., Yin, Y. and Gao, H. (2024), FMGDN: Flexible multi-grained dilation network empowered multimedia image inpainting for electronic consumer (pdf), in IEEE Transactions on Consumer Electronics, DOI https://doi.org/10.1109/TCE.2024.3386773.
___________________________________________________________________________________________________________________
NOTES on three-dimensional VOLUME DATA RECONSTRUCTION
from scans and CLASSIFICATION OF MATERIALS AND OBJECTS
extracted from volume data, ~1650 pages, prepared by
Bernd Hamann, to appear.
The notes concern the processing of data generated by imaging and scanning technology for the analysis of volume data in a non-destructive way. In a first step, usually two-dimensional images are produced by having rays pass through a three-dimensional region of space that contains objects to be identified. Reconstruction uses the two-dimensional images to estimate object and material properties in three-dimensional space. In a second step, the reconstructed volume is processed by a multitude of algorithms to determine spatial regions that represent the distinct materials. Based on known material properties of specific material classes to be found in the volume data, data segmentation and statistical analysis methods are used to classify the materials and objects extracted from the volume.
Volume data reconstruction is understood as the numerical approximation of material density values for voxels inside a region in three-dimensional space that contains objects with specific material properties. Generally, two-dimensional digital images define the input to the reconstruction process. A radiation source behind the region in space generates rays that pass through the objects of interest, producing two-dimensional images (projections) in front of the region. Material density can be estimated from the images by considering the applicable physical laws defining the imaging process. Various numerical methods are used for reconstruction, including methods from radiative heat transfer, image processing, geometric computing, approximation theory, optimization, linear algebra and volume rendering. The notes summarize important aspects of the volume scanning procedure, relevant physical laws and computational approaches used in volume reconstruction algorithms.
Classification of materials and objects is based on the extraction of regions in reconstructed volume data, represented by voxels with associated property values. The properties of these regions (segments) are compared with stored characteristics of materials to be recognized in the volume data. Extraction of segments is a complex process that involves the training with a multitude of sample data. Training makes it possible to learn the material characteristics from the samples of different classes, thus enabling later material recognition. Material and object classification uses techniques from several areas, including image processing, statistics, probability theory, algebra, linear algebra, logic, differential geometry, analytical geometry, geometric algebra, geometric analysis, vector calculus, differential equations, spectral and wavelet data analysis and multiresolution methods. Effective and efficient material- and object-specific application of these methods depends on value optimization of method parameters to obtain classification results with high degrees of certainty.
Illustrations and numerical examples are used frequently in the notes. They explain geometrical and mathematical concepts used in the data processing algorithms.
The notes contain errors and should be understood merely as a high-level summary of problems, solution approaches and algorithms related to volume data processing.