A method for reducing noise in flouroscopic images by detection of the traits in the curvelet domain, matching of the detected traits and time-domain filtering adapted to the type of coefficients associated with a trait. The method uses the discrete curvelet transforms of the images from a sequence. The denoised coefficients are detected, then transmitted to a step for locally matching the traits, the matched data subsequently undergo a step for time-domain filtering allowing the movement of the objects to be preserved in the image and a step for a 2D IDCT (Inverse Discrete Curvelet Transform) in order to produce the final image.