基于超复视域注意模型的视频分割算法

Video Segmentation Algorithm Based on Hypercomplex Visual Attention Model

提出一种基于超复视域注意模型的视频分割算法,无需事先针对特定类型的目标进行训练。通过构造超复视域注意帧图像,对超复视域注意帧图像计算相位相关实现运动建模,利用条件随机场对视域注意模型、颜色模型以及邻域关系模型进行约束求解,获得分割结果。采用不同的视频数据对该算法的有效性进行测试,并与其他分割算法的结果进行比较。实验结果表明,该算法的分割错误率较低。

Automatically segmenting out non-specific objects from moving background is a difficult problem. A method based on hypercomplex visual attention model for video segmentation is proposed, which does not require training specific class of objects. The algorithm constructs hypercomplex visual attention frames to model motion via computing phase correlation. Conditional random fields are used to＋ constrain visual attention models, color models and neighboring relationship models to obtain segmentation results. Experimental results demonstrate the validity of proposed algorithm, and results show that the error rate is lower compared with other algorithms by using different video data.