Removing blue screen background under non-uniform illumination

In this paper, we develop a system to remove the background of an image captured in a blue screen environment under non-uniform illumination. We first collected some images as the training data and applied a watershed segmentation to extract a number of coherent regions from these images. With these extracted regions, two mixed Gaussian models for the foreground and background are trained using an expectation-maximization (EM) algorithm. With the trained models, for each input image we first segment the image using the watershed segmentation algorithm and then predict the associated foreground/background label for each extracted region. Finally, the regions labeled as background are removed. Some experiments were conducted and the results showed that the proposed system performs better than traditional chroma keying technique.