| Popis: |
Change detection is a popular and challenging research field in which the changes between two images of the same location collected at different times are detected. The crux of the problem is to detect only significant changes, based on application goals, while suppressing insignificant changes. Insignificant changes may be related to a variety of items including illumination changes, natural seasonal changes, atmospheric conditions and image registration errors.This dissertation involves further investigation into ratio-based change detection. Standard ratio-based change detection methods use a single ratio along with a threshold and its reciprocal to detect changes in both tails of the ratio distribution. Ratio-based methods show unique ability among change detection algorithms from published literature to detect challenging changes. However, the ratio-based methods in literature also exhibit problems detecting certain types of changes and suffer from high false alarm rates.A multi-ratio fusion framework for robust change detection, based on ratio-based change detection, is proposed and tested in this dissertation. A method called dual ratio (DR) change detection is developed featuring two ratios coupled with adaptive thresholds to maximize detected changes and minimize false alarms. The use of two ratios is shown to outperform the single ratio case when the means of the image pairs are not equal.A multi-ratio (MR) change detection method is developed building upon the DR method by including negative imagery to produce four ratios with adaptive thresholds. Inclusion of negative imagery is shown to improve detection sensitivity and to boost detection performance in certain target and background cases.A multi-ratio fusion (MRF) change detection technique further expands the algorithmic concepts in DR and MR by fusing together the ratio outputs to maximize detections and minimize false alarms. In the fusion algorithm, detections must be verified by two or more ratios in order to be classified as a true changed pixel. The proposed methods are tested with synthetically generated test imagery and real datasets with results compared to other methods found in literature. MRF produces excellent change detection results that exhibit up to a 22% performance improvement over other methods from literature at low false alarm rates. |
