In large-scale high-power laser devices, the laser energy density and the system operating speed are mainly limited by the damage growth of the terminal optical element. An adaptive differential window filtering method based on local signal-to-noise ratio is proposed in order to detect the damage state of the element efficiently and accurately. By designing a window function which acts on pixels, the algorithm can judge the target point or background point according to the pixel value and the values of its neighborhood point. And then the segmentation of the seed image is completed. Finally, the damage segmentation is completed by the growth of the seed region. In order to verify the effectiveness of the algorithm, we built an on-line detection simulation platform to obtain the damaged sample image, and the new method mentioned above is used to process the image. The results show that the average recognition rate of the method is above 99% for the damage points with the diameter above 50 μm, which meets the requirement of high-power laser drive system for the detection of small damage. Because the algorithm does not require setting the threshold of the seed image based on experience, it is more automated than the existing local signal-to-noise ratio algorithm.