A refinement on detection in cell counting

Cell counting is an important task and it's difficult for the naked eyes to keep counting cells one by one for a long time. In this paper, we use multiple random forests (RFs) to detect cells and propose a post-processing to refine the detection result. Each RF generates a different detection result and averaging all detection results can decrease the count error. For some false detected cells approved by a few RFs, they have low confidence in the averaged detection result. To improve the count accuracy further, we propose a post-processing to remove the low-confident detections and enhance the high-confident detections. We validate our proposed method on two cell datasets and prove that our proposed method can refine the count result and improve the performance.