Performance comparison of the Teknomo-Fernandez algorithm on the RGB and HSV colour spaces

Segmentation of the foreground objects is the primary step in many video analysis applications. The accuracy of the segmentation is dependent on an accurate background image that is used for background subtraction. The Teknomo-Fernandez (TF) algorithm is an efficient algorithm that quickly generates a good background image. A previous study showed the extendibility of the TF algorithm to higher number of frames per tournament, with the original 3 frames TF_3L to be the most efficient and best configuration for actual implementation. In this study, we examine the performance of the TF algorithm on both RGB and HSV colour spaces using the TF_{3, 4} configuration and the Wallflower dataset. A simple background subtraction with threshold is implemented. The performances are measured numerically using the number of false negative and false positive pixel count against the provided ideal foreground image. The results show that the TF algorithm implemented using both RGB and HSV generates accurate background images in a wide range of video settings. The HSV implementation exhibits higher accuracies than the RGB implementation for majority of the test videos with the cost of an increase in processing time.