Adaptive Shadows Detection Algorithm Based on Gaussian Mixture Model
Yu-jiao XiaHou, Shengrong Gong · 2008
This paper proposed an adaptive shadows detection algorithm based on Gaussian Mixture Model to improve the performance of video object segmentation. This method takes advantage of luminance weight to model the background of the image and obtains a primary segmentation in CIE Luv color space. In this way, it improves the real-time ability of detection. It also becomes more efficient, comparing with the existing shadow detection algorithms which often need to set the threshold manually or get them through a training process. By using the Gaussian distribution, it is able to realize an adaptive shadow detection. At same time, the authors deal with the noise or the aim points uneven distribution by using horizontal filling and vertical filling. It improves the accuracy of segmentation. The experimental results have shown that this method achieves adaptive shadows detection and has strong robustness, high segmentation accuracy.