We propose a novel approach for the photon shot noise reduction in image sensors inspired from an M-transform method. In our proposed method, the photon shot noise is transfered to small amplitude random signals. Then the small amplitude random signals are removed by the ε filter. The Sobel method was used for the edge detection and the edge was preserved by not applying the ε filter into the edge. After the noise reduction, the signals are restored to the image by a reverse M-transform. We compared our method with the smoothing filter, the median filter and the Wiener filter. In the photon shot noise according to the Poisson distribution, we changed the incidence number of photons and created the noise image by the simulation. In cases where the number of the maximum incidence photons was 128 pieces (illumination of about 2 lux), the proposal technique indicated the PSNR (Peak-Signal to Noise Ratio) value better than the Wiener filter. Furthermore, the proposal technique confirmed better than the smoothing filter subjectively according to the effect of edge preservation also under dark environment. As the result, our method is confirmed effective for the photon shot noise reduction.