SEGMENTING FOCUSED OBJECTS IN COMPLEX VISUAL IMAGES

In this paper we aim at the segmentation of focused objects in complex visual images that contain foreground and background objects. The pro posed approach is an unsupervised edge-based segmentation that relies on the measurement of defocus (blur) for object edges in an image. The amount of defocus at an edge pixel is evaluated using the moment-pres erving principle. Edges of focused objects generate small. amount of d efocus, whereas edges of background objects yield large amount of defo cus. Following an edge-linking procedure, the regions of focused objec ts can be bounded by closed boundaries. Experiments on a variety of vi sual images have shown correct segmentation, regardless of the complex ity of foreground and background in images. (C) 1998 Elsevier Science B.V. All rights reserved.