基于三维椭圆编码本模型的视频前景分割

Foreground Segmentation Based on 3D Ellipses Codebook Model

针对基于编码本模型的前景分割算法处理颜色较暗的区域存在较高的不稳定性,并且对于克服光照变化和前景目标阴影的参数难以调节这两个缺点,提出了一种基于三维椭圆编码本模型的前景分割算法。该算法在RGB颜色空间中建立三维椭圆模型,根据光照环境调节控制光照变化的椭圆长轴和控制噪声影响的椭圆短轴消除阴影和光照变化带来的干扰;依据像素点的颜色值是否位于该三维椭圆外,则将其判为前景目标;同时自适应更新三维椭圆的中心位置以适应背景的光照变化。对比实验表明,该改进方法在处理视频中的较暗区域效果理想,并且参数直观明确易调整,对于准周期性变化的动态背景也能达到较好的分割效果。

The video segmentation method based on the codebook model has high instability in processing dark regions,and is hard to adjust the parameters for overcoming the illumination change and shadows of foreground objects,to solve these two shortcomings a novel foreground segmentation based on 3Dellipses codebook model is proposed.This method establishes a 3Dellipses model in RGB color space,and to remove the interference of shadows and illumination change by regulating the major axis which controlls illumination change and the minor axis which controlls noise through illumination conditions;if the color vector of a pixel locates outside of 3Dellipses,it is judged as foreground object;simultaneously adaptively update the central of 3Dellipses to adapt the illumination change.Comparative experimental results demonstrate that this method has high stability in processing dark regions,and the parameters which are easily to adjust are intuitive and explicit.To deal with the video which has periodic-like motion,it can achieve better segmentation result.