The effect of color space and block size on foreground detection

Background subtraction techniques, in order to identify moving objects, are commonly used in computer vision applications. This is still a challenging problem especially when there is a non-stationary background. Two most significant tasks of generic background subtraction techniques are the background modeling, which determines how background will be represented and detect the foreground region, which significantly differ from the background model. The problem with which features will be represented in both background modeling and foreground detection is an important research topic. The purpose of this study is to determine the most effective color space, region block size and distance metric for foreground detection regardless of the background model. As a performance metric, the area under the Receiver Operating Characteristic (ROC) curve is used. Tests were performed over 9 different videos which have non-static background in I2R dataset.