Hybrid image segmentation for Earth remote sensing data analysis

Image segmentation is a partitioning of an image into constituent parts using image attributes such as pixel intensity, spectral values, and/or textural properties. Image segmentation produces an image representation in terms of edges and regions of various shapes and interrelationships. It is a key step in several approaches to image compression and image analysis. The author has devised a hybrid image segmentation approach that combines region growing and boundary detection. The core of this image segmentation approach is an iterative parallel region growing algorithm that the author has developed over the past several years. The question of where to stop the region growing process is solved by not allowing the region growing process to grow regions past boundaries defined by a boundary detection algorithm. The author has found an edge detector based on an optimal difference recursive filter to be most suitable for this boundary detection. This edge detector provides highly localized edge boundaries and is relatively insensitive to noise. It also provides a convenient threshold parameter through which an application appropriate edge density can be selected.