Electron multiplier CCD (EMCCD) can po tentially count a single photon because of the very low detection noise.But the gray values of adjacent pixels are highly correlated with each other in pho ton grayscale images obtained by EMCCD (coorelation coefficients are larger than 0.72).This feature shows that the number and location of photon signals will ha ve a greater distortion if the images are processed based on the threshold algor ithm.In addition,the number of photon signals isolated from each other in an ima ge is much smaller than that of pixels under the situation of ultraweak luminesc ence,e.g.Therefore,a new algorithm was developed based on watershed algorithm by treating one isolated high-value region as the response of one photon.This new algorithm gives baseline estimation synchron ously,and takes the characteristics of group pixels into account rather than a s ingle pixel.The results show that the new algorithm is more reliable and stable due to reducing the influence of unsureness of optimal threshold value and prote cting the location information of photons better.The sensitivity and reliability of this new algorithm are also tested and confirmed with EMCCD imaging of mung bean sprouts,even the signals are as weak as 146photons/image.