The flash presence is very common in digital videos, mainly in television journal videos. When a flash occurs, an increase of the luminosity in a few frames is produced. When we calculate a dissimilarity measure, we can see that this measure is very high in the frames affected by a flash. In fact, the presence of flashes often disturbs the identification of video transitions. In this work, we transform the video into a 2D image in which the flash is represented by a vertical light line. Considering this image, we propose two methods for flash detection without
taking into account dissimilarity measures. The first one
considers a filtering using white top-hat by reconstruction and the second one considers a filtering of the max tree calculated from statistical measures of each frame (or frame sub-samplings).