| Summary: | In this thesis, a sonar mine detection system is presented. The three stages of the system, image enhancement, edge detection and data compression are analyzed. For each stage, different approaches are studied and the appropriate algorithm is chosen.
For image enhancement, a cascaded morphological enhancement algorithm is developed. Our algorithm is shown to preserve high-detail regions, including edges surrounding features of concern (suspect mines) and their corresponding shadows. This is accomplished without introducing any blurring, which is an inherent side-effect of conventional enhancement algorithms.
For edge detection, amongst various methods, we have found the two dimensional morphological edge detector, the Alpha-Trimmed Morphological (ATM) filter, to be immune to noise, a major problem encountered in sonar imagery, and consequently this filter has outperformed the other conventional edge detectors.
At the compression stage, three algorithms are examined. The Vector Quantization (VQ) method, using a new codebook generation algorithm, which we have developed, is found to yield very high quality reconstructed images. The decompressed images have their high-detail regions well preserved, and furthermore the low-detail regions are minimally distorted, to the extent that the reconstructed images are virtually indistinguishable from the original images. With the new compression algorithm, a compression ratio of 1:10, or equivalently 0.8 bits per pixel (bpp) is achieved. Furthermore, the processing time for codebook generation is greatly shortened. === Applied Science, Faculty of === Electrical and Computer Engineering, Department of === Graduate
|
|---|
