Abstract: Foreground extraction of moving object is the foundation for further analysis.An almost complete object foreground can be obtained by RPCA (Robust Principal Component Analysis) decomposition.However,this approach is sensitive to illumination change.The robustness of RPCA-based foreground extraction approach can be increased by the fact that a,b channels of Lab color space are not sensitive to illumination change.Initially,the sparse foregrounds of L,a,b channels are calculated based on RPCA decomposition respectively.Then Ostu thresholds are employed for binary foreground segmentation of each channel and seed filling technology is utilized for fusing different foregrounds.Finally,the accurate moving object foregrounds are extracted after improving the fusion results with morphology filtering.The experimental results indicate that the proposed method can accurately extract object foreground under complex environments while handling illumination change effectively.

Key words: Robust principal component analysis,Moving object foreground extraction,Lab color space,Seed filling