Degree Name

Doctor of Philosophy


School of Electrical, Computer and Telecommunications Engineering


Effective through-the-wall surveillance (TWS) technologies have attracted major research interests and are considered a valuable tool for protecting and safeguarding a country from terrorism and crime. Besides reconnaissance operations, the ability to sense through visually opaque materials is also useful in search-and-rescue missions to locate victims during disastrous tragedies. While there are reported successes in through-the-wall radar imaging (TWRI) systems, acquired images tend to be cluttered, blurred, shifted and smeared. Thus, numerous improvements in the image processing domain can be envisioned for future systems, which allow the successful extraction of relevant information for human analysis and machine perception.

This dissertation has yielded innovative solutions that improve the capabilities of TWS systems, thereby reducing the complexity associated with detecting, tracking, classifying, and identifying objects of interest behind walls. More specifically, advanced image processing algorithms that enhance the sensed images are developed. The algorithms proposed consist of image segmentation and image fusion techniques, which act as a pre-processing step, either individually or collectively, for target detection, automatic target recognition, or information extraction from through-the-wall radar (TWR) images. Commonly used image segmentation techniques, such as the entropy-based segmentation, between-class variance thresholding and K-means clustering are first investigated to distinguish between target and clutter regions in the acquired images. A Gaussian-Rayleigh mixture model is then developed to improve the segmentation results. A fuzzy logic-based image fusion method is proposed to combine multiple images of the same scene acquired from different view angles, or at different polarizations. A probabilistic fuzzy fusion method is then developed, which alleviates the need for the ad-hoc design of membership functions for fuzzy fusion; the fuzzy membership functions are learned automatically from the input images using a Gaussian-Rayleigh mixture model. The proposed methods are extensively evaluated on real TWR images. Experimental results on both polarimetric and multi-view images show that the developed segmentation and fusion algorithms successfully enhance target information while suppressing clutter in the images, which yields a higher image quality and improves target detection.