关键词: 图像分割, 概率分布, 轮廓提取, 目标检测, 无参密度估计

Abstract: Focusing on object contour extraction technology that had the most significant impact on the process of image analysis, an efficient method was proposed for object contour extraction based on probability distribution. First the foreground object was detected and its hue feature was used to build an object probability model. With this model, the probability distribution image was calculated on the whole image for the object. Finally, the contour was obtained from the probability distribution image directly. After testing the method using 2330 continuous images containing moving object, the results suggest that the method can work efficiently. It takes 20ms to 30ms to process a true color image of 640×480 pixels, and the contours extracted are very clear and accurate.

Key words: image segmentation, probability distribution, contour extraction, object detection, null-parameter density estimate