Three-dimensional image processing and recognition
Date of Completion
January 2006
Keywords
Engineering, Electronics and Electrical
Degree
Ph.D.
Abstract
Two research areas are investigated in this dissertation. One is pattern recognition with three-dimensional object information. The other is compression of three-dimensional ray information in integral imaging. ^ The ability to detect and identify unknown objects in input scenes and label them as one of hypothesized classes has been investigated for the past decades. The discrimination capability of recognition systems is often challenged by arbitrary distortion and high intra-class variation of object patterns. ^ This dissertation addresses several mathematical and computational methods for the recognition and classification by means of three-dimensional information. Three-dimensional information is acquired by various imaging techniques such as integral imaging, photon counting integral imaging, x-ray volumetric imaging, and computational holography. The following is a summary of recognition techniques in this dissertation: (1) nonlinear matched filtering and photon counting linear discriminant analysis with photon counting integral imaging for automatic target recognition, (2) thickness feature evaluation method to identify filamentous structure of microorganisms, (3) three-dimensional Gabor-based wavelets and three-dimensional dynamic link association (graph matching) for the analysis and recognition of volume objects, (4) complex morphology-based recognition and automatic feature selection for the identification of microorganisms, and (5) principal component analysis followed by mixture discriminant analysis to utilize multi-spectral information. ^ A low level of photons is needed for object recognition tasks using the proposed techniques in photon counting integral imaging. The Gabor-based wavelets decompose geometrical shape and extract feature vectors. By evaluating thickness features, filamentous microorganisms with different thickness are discriminated. The graph matching technique with Gabor-features is shown to be robust to translation, rotation, and distortion of volume objects and complex morphology of microorganisms. Multi-spectral information is embedded into the recognition system to classify color objects. ^ For the compression of three-dimensional ray information in integral imaging, two compression schemes are addressed: (1) the hybrid compression scheme composed of vector quantization, principal component analysis, and rounding-off, and (2) MPEG-2 compression with three scanning topologies of elemental images. Both of compression methods are shown to be superior to JPEG for high compression ratios.^
Recommended Citation
Yeom, Seokwon, "Three-dimensional image processing and recognition" (2006). Doctoral Dissertations. AAI3236157.
https://digitalcommons.lib.uconn.edu/dissertations/AAI3236157