Real-time snowfall noise elimination

We propose a new method that removes snowfall noise from successive images recorded by a TV camera. This method utilizes the size and velocity of a moving object. The intensity of a pixel changes when an object passes over the pixel. Snow particles pass a pixel very quickly due to their small size. Therefore, after holding the odd number of video frames, they are sorted by intensity in each pixel independently. Then we can obtain the background intensity of the pixel by taking the median intensity from them. This was confirmed by experiments. We also proposed a stochastic model of this algorithm. By comparison between a measurement value and the model, we obtained the agreement with each other. Furthermore, we consider the effect for the other moving objects. Experimental results are also shown.