Research Seminars

A Segmentation Algorithm for Extended Target in Complex Environments
Prof. Zhang Qiheng, Director of the Institute of Optics and Electronics, Chinese Academy of Sciences
Tuesday 12th November 2002

Target segmentation is a key technique in automatic recognition and tracking. When the distance between target and camera is becoming gradually closer, the weak and small characteristic of the target becomes larger and extended. Those targets, with size exceeding 100X100 pixels and with area smaller than a third of the field of view, are called extended targets. Their large size, complex shape and inhomogeneous surface make difficult the segmentation of extended targets in simple environments. When such extended targets are located in natural complex environments, the problem becomes awfully difficult. Conventional methods such as homogeneous characteristic segmentation and thresholding can not be adopted. Currently, only few works about segmenting complex targets in complex environment can be found. Most reported techniques assume targets with homogeneous gray level.

In this talk a technique for the segmentation of extended targets from complex environment is presented. The method is based on knowledge, fractal and mathematical morphology. Knowledge based in the assumption that addressed targets are man-made objects in front of natural background is used together with the fact that the fractal dimension of man-made objects and natural background are very different. Multi-features of the target, such as boundaries area and texture are utilized in a morphological approach to eliminate the background.