A Novel Approach to the Noise Removal and Detection of Small Objects with Low Contrast

An effective approach for detection of small objects with low contrast is proposed. In our work, small moving objects are detected in an image sequence captured from a video camera. The detection system includes two modules, namely region of interest (ROI) locating and contour extraction. In the former module, image-differencing technique is employed on consecutive images to generate rough candidates of objects appearing in the images. Next, a novel neighboring encoding technique is devised to effectively remove noise that usually severely affect the performance of detection. The noise-removed candidates are then bounded by minimum enclosing rectangles to obtain ROI. However, the generated results provide insufficient information characterizing object contours. Traditional contour extraction methods cannot obtain satisfactory results under this circumstance. We adopt the watershed algorithm incorporating with region-matching technique to obtain accurate object contours. Experimental results demonstrate the feasibility and effectiveness of proposed approach.