Stochastic Resonance and Mean Shift Filtering for Detecting Weak Features in Noisy Images

Abstract

Stochastic Resonance has been shown to occur in many biological, physical and geological systems, resulting in the boosting of weak signals to make them detectable. In the image processing domain, narrow regions, small features and low-contrast or subtle edges, especially in noisy images, correspond to such weak signals. We show, both mathematically and empirically, that stochastic resonance occurs and may be exploited in the detection, extraction and analysis of such features. These mathematical results are confirmed by simulation studies. Finally, results on standard images such as cameraman, boats, lena, etc. demonstrate that several subtle features lost by the application of robust techniques such as Mean Shift filter are recovered by stochastic resonance. These results reconfirm the mathematical and simulation findings.