Record:   Prev Next
Author Wang, Mao
Title Active image registration and recognition
Descript 135 p
Note Source: Dissertation Abstracts International, Volume: 56-08, Section: B, page: 4510
Co-Directors: Bruce Walcott; Larry Hassebrook
Thesis (Ph.D.)--University of Kentucky, 1995
Image feature matching is the key to image registration and recognition. In this dissertation, an active image feature matching technique is developed, which incorporates both local and global information in the matching process to achieve a global optimal goodness-of-match. In addition, since this technique is contour-based, it is particularly useful in situations of multisensor images where images have different gray level characteristics but contours representing region boundaries are preserved. First, an optimal active contour (snake) is used to reduce the 2D object of interest to a 1D contour feature string. This snake has the capability to extract accurate information about an object's corners that contains significant discriminant information. High performance is achieved by dividing the energy optimization process into multiple stages. The first stage was designed to optimize the convergence speed in order to allow the snake to quickly approach the optimal state. The second stage was devoted to snake refinement and to the local optimization of energy thereby driving the snake to the quasi-optimal state. The third stage uses the Bellman optimality principle to fine tune the snake to the global optimal state. This three-stage scheme optimized both performance and speed of the snake. After the objects to be matched were reduced to two feature vector strings, dynamic feature matching (DFM) was used to match these strings. DFM matched the two feature strings in a global optimal way by using both the Bellman optimality principle as well as a back-propagation neural network. Also, a hidden Markov model for dynamic feature matching was derived. This model shows that the dynamic feature matching is optimal in the sense of maximum likelihood. An active image registration system was then introduced using active feature matching to obtain a partial disparity map from which a full disparity map was estimated using regularization. This system is tested on a sequence of MR functional brain images. Results show that the brain activation map obtained from registered images was significantly improved when compared to nonregistered images. Finally, an active image recognition system is implemented based on active feature matching. This system is applied to aircraft images and results indicate that the active recognition system has superior distortion tolerance over the correlation based system. It maintains good performance over a wide range of distortion. This tolerance to distortion is due to its "active" nature. In other words, to some extent, it mimicks human vision by dynamically adjusting the matching path so that the differences due to perspective distortion was minimized
School code: 0102
Host Item Dissertation Abstracts International 56-08B
Subject Engineering, Electronics and Electrical
Artificial Intelligence
Alt Author University of Kentucky
Record:   Prev Next