An reflection normalization method based on parameters training

To solve the light problem encountered in video segmentation and video analysis, a new normalization method based on parameters training is proposed in the paper. In the method, a multi-dimensional Gaussian mixture model is given firstly to extract background and acquire the moving targets. Then, the reflection areas in the targets are identified, and according to the continuity of continuous frames, the relationship parameters for reflection areas normalization are obtained in training procedure. Thirdly, the reflection areas in the targets are recovered by the parameters. The experimental results show that the most reflection areas could be recovered by the method and do good to the next video analysis.