Three dimensional distorted noisy pattern recognition using optimum nonlinear filters
Date of Completion
Engineering, Electronics and Electrical
The presence of noise including disjoint background noise and additive noise between the target and the sensor degrades the recognition or classification performance of pattern recognition systems. A novel methodology for optimum distortion-tolerant detection has been presented. This technique takes into account the disjoint background noise, additive noise, false objects, and true class nontraining targets. The criterion used to design the filter is to minimize filter output energy due to the input scene and noise. It is designed and implemented to detect and classify true class targets under such conditions. The statistical detection performance of the designed distortion-tolerant optimum nonlinear filter has been tested by computer simulations. I have extended the recognition algorithm into a 3D form to detect and recognize 3D LADAR range objects in the 3D LADAR range scene, and conducted the performance tests. ^ The second contribution of this dissertation is the computational acquisition of 3D information to apply the implemented filters to 3D pattern recognition I have proposed a novel computational integral imaging reconstruction method that allows full profile information of the 3D scene. The quality of the reconstructed integral images has been further improved by a new computational moving array lenslet technique with a normalization process. ^ The third contribution of this dissertation is the 3D fusion of images from the output of multiple sensors. To accomplish this technique, I have combined the integral imaging system with multi-wavelength images, such as multi-wavelength digital holographic images and near infrared images, and visual band images. These fused or non-fused 3D images can be applied to the 3D pattern recognition system.^
Hong, Seung-Hyun, "Three dimensional distorted noisy pattern recognition using optimum nonlinear filters" (2005). Doctoral Dissertations. AAI3193721.