Abstract: General digital cameras, camcorders as well as the BOYS camera specially developed for the lightning observation, are all important tools for lightning research. They are used to obtain the basic data for understanding geometry features of channels and main development processes of lightning. For a long time, extracting lightning channel coordinates from digital images is based on manual processing methods with the relatively low efficiency. With the development of the photoelectric techniques, more and more advanced optical devices are used in lightning observation, such as high-speed cameras and the Automatic Lightning Processing Feature Observation System (ALPS), and data obtained become much richer. How to automatically process these data and improve the efficiency of data extraction and analysis is an urgent need to be addressed.Considering the complexity of lightning discharges and various characteristics of the lightning channel, only one algorithm is not enough to obtain a satisfying recognition result in all situations. Therefore, 3 common threshold methods are applied jointly in the lightning channel recognition. Firstly, the impact of the uneven illumination is eliminated by subtracting the background and the contrast of the image is enhanced. Secondly, global adaptive threshold method, local adaptive threshold method or adaptive Canny operator method is applied for edge detection. And then, morphological and thinning processes are carried out to get the lightning channel represented by the continuous sequence of pixels. Considering different characteristics of the lightning channel digital image, selecting appropriate algorithm can ensure getting a clear edge information even including weak edges, guaranteeing a good recognition effect finally.Through experiments, it can be concluded that for the lightning channel with simple structure and relatively uniform brightness, all algorithms mentioned above can get a good recognition result, among which the global adaptive threshold method is simpler and more efficient. Local adaptive threshold method can calculate the threshold for different images universally. And for the low-contrast images with a smooth background, using adaptive Canny operator method can achieve a satisfactory recognition result.