Invariant Color Model-Based Shadow Removal in Traffic Image and a New Metric for Evaluating the Performance of Shadow Removal Methods

摘要

To track objects in a traffic image sequence, objects must be extracted first. Background differencing is frequently used to extract objects. When objects are extracted, it is quite possible that shadows are included. With shadow it is not easy to do precise tracking. Thus shadows need to be removed. To do this, we proposed invariant color-based shadow removal method. Many shadow removal methods were proposed. To compare the quality of methods, several metrics were suggested. However, they suffer from inconsistency where qualitative and quantitative results do not coincide. In this paper, we proposed a new metric having such consistency.