Engineering Commons^™

Color Image Segmentation, Kimberley A. Mccrae

Theses and Dissertations

The most difficult stage of automated target recognition ATR is segmentation. Current AFIT segmentation problems include faces and tactical targets previous efforts to segment these objects have used intensity and motion cues. This thesis develops a color preprocessing scheme to be used with the other segmentation techniques. A neural network is trained to identify the color of a desired object, eliminating all but that color from the scene. Gabor correlations and 2D wavelet transformations will be performed on stationary images and 3D wavelet transforms on multispectral data will incorporate color and motion detection into the machine visual system. The thesis …

Simplification Of The Generalized Adaptive Neural Filter And Comparative Studies With Other Nonlinear Filters, Henry Steven Hanek

Theses

Recently, a new class of adaptive filters called Generalized Adaptive Neural Filters (GANFs) has emerged. They share many characteristics in common with stack filters, include all stack filters as a subset. The GANFs allow a very efficient hardware implementation once they are trained. However, there are some problems associated with GANFs. Three of these arc slow training speeds and the difficulty in choosing a filter structure and neural operator.

This thesis begins with a tutorial on filtering and traces the GANF development up through its origin -- the stack filter. After the GANF is covered in reasonable depth, its use …

Evaluation Of Collision Properties Of Spheres Using High-Speed Video Analysis, Jian Yu

Theses

Experimental evaluation of the collision properties of spheres is performed using video image analysis techniques. A high-speed Kodak EktaPro1000 video camera is utilized to record a collision sequence between two spheres at 1000 frames/sec, and then the images are analyzed to calculate three dimensional translation and rotation before and after the collision. These quantities are used to compute the collision properties for a pair of one inch nylon spheres, i.e. the coefficient of friction, and the coefficients of normal and tangential restitution. The focus of the thesis is on image analysis techniques that provide high accuracy results even though the …

Detection Of Planar Facets In Noisy Range Images, Ajey S. Atre

Theses

Segmentation of the image is one of the major tasks of a machine vision system designed for constructing a three-dimensional representation of the object being imaged. A robust approach for segmenting planar surfaces from range images is presented in this paper. An algorithm based on clustering through fuzzy covariance matrices, which has been proposed by Gustafson and Kessel is considered for planar segmentation. However this algorithm performs poorly if the data is noisy, which is usually the case in real life applications. In order to handle noisy data, a robust modification, based on the "noise clustering" concept, is introduced to …

Optical Image Analysis Using Fractal Techniques, Samuel Peter Kozaitis, Harold Gregory Andrews, Wesley E. Foor

Electrical Engineering and Computer Science Faculty Publications

Using an optical technique, we classified images of natural terrain based on their fractal dimension. We calculated the fractal dimension from an optically generated power spectrum obtained with a magneto-optic spatial light modulator (SLM). By using the fractal dimension to classify images of natural terrain, our post processing was simpler that when a ring-wedge detector was used.

Feature-Based Correlation Filters For Object Recognition, Samuel Peter Kozaitis, Wesley E. Foor

Electrical Engineering and Computer Science Faculty Publications

Using an optical correlator, we experimentally evaluated a binary phase-only filter (BPOF) designed to recognize objects not in the training set used to design the filter. Such a filter is essential for recognizing objects from actual sensors. We used an approach that is as descriptive as a BPOF yet robust to object and background variations of an unknown or nonrepeatable type. We generated our filter by comparing the values of spatial frequencies of a training set. Our filter was easily calculated and offered potentially superior performance to other correlation filters.

A Direct Method For Surface Structure Recovering Based On Uoff, Ping Lin

Theses

The unified optical flow field (UOFF) theory which can be used for estimating motion and recovering surface structure was recently established in [9, 10]. The direct method developed in [2, 3, 4, 6, 7] does not need to explicitly solve the optical flow field and to find feature correspondence. Based on the UOFF, a direct method in space domain is developed to reconstruct the curved surface structure characterized by an Nth degree polynomial equation from a pair of stereo images. The initial work on this new method was reported in [8, 11].

In this study, I basically work on simulation …