Abstract:

Segmentation is the first step towards the target location and recognition in video sequences. This task becomes difficult when the background illumination changes. A Bayesian learning method was applied into video segmentation. The constantly changing background was modeled at the pixel level. The histogram colors and co-occurrence vectors were calculated. The feature vector for each pixel was represented with a discrete probability distribution function. A Bayesian learning method was used to obtain these probability distribution functions from the video image inputs. Experiment results indicate that the proposed approach is able to learn a complex background of which the illumination changes gradually or suddenly.

Key words: video image segmentation, Bayesian learning, complex background modeling

摘要：

视频图像分割是视频目标定位和识别的基础,如果背景中光线变化,那么将会给分割带来很大的影响。文中利用贝叶斯学习方法进行视频图像分割,在每个象素点处对不断变化的背景建模,计算每个象素点处的颜色直方图,用这些直方图来表示该象素点处特征向量的概率分布,然后用贝叶斯学习方法来判断,在光线缓慢或者突然变化的时候,每个象素点是属于前景还是属于背景。

关键词: 视频图像分割, 贝叶斯学习, 复杂背景模型