Algorithms for regions and boundary detection in images are important for biomedical data analysis both for research and diagnosis. In this talk I will present algorithms for edge detection and segmentation that use a multiscale approach to overcome problems of noise and poor contrast, which are common in biomedical images. Our edge detection algorithm is based on a fast method for computing filter responses at different lengths and orientations and on a recursive decision process which selects the desired responses from all scales. Our image segmentation algorithm combines intensity and texture cues and outputs a full hierarchy of segments. These tools were applied to a variety of biomedical images acquired by light and electron microscopes and by Magnetic Resonance Scanners, and tested on a variety of applications including the identification of cell structures, tissues, and abnormal lesions.
Back to Workshop IV: Search and Knowledge Building for Biological Datasets