Image segmentation is generally the first task in any automated image
understanding application, such as autonomous vehicle navigation, object
recognition, photointerpretation, etc. All subsequent tasks, such as
feature extraction, object detection, and object recognition, rely
heavily on the quality of segmentation. One of the fundamental
weaknesses of current image segmentation algorithms is their inability
to adapt the segmentation process as real-world changes are reflected in
the image. Only after numerous modifications to an algorithm's control
parameters can any current image segmentation technique be used to
handle the diversity of images encountered in real-world applications.
Genetic Learning for Adaptive Image Segmentation presents the first
closed-loop image segmentation system that incorporates genetic and
other algorithms to adapt the segmentation process to changes in image
characteristics caused by variable environmental conditions, such as
time of day, time of year, weather, etc. Image segmentation performance
is evaluated using multiple measures of segmentation quality. These
quality measures include global characteristics of the entire image as
well as local features of individual object regions in the image.
This adaptive image segmentation system provides continuous adaptation
to normal environmental variations, exhibits learning capabilities, and
provides robust performance when interacting with a dynamic environment.
This research is directed towards adapting the performance of a well
known existing segmentation algorithm (Phoenix) across a wide variety of
environmental conditions which cause changes in the image
characteristics. The book presents a large number of experimental
results and compares performance with standard techniques used in
computer vision for both consistency and quality of segmentation
results. These results demonstrate, (a) the ability to adapt the
segmentation performance in both indoor and outdoor color imagery, and
(b) that learning from experience can be used to improve the
segmentation performance over time.
