Abstract: An image output by a camera can contain a fault such that the signal output by the camera is not constant for a uniform background. To correct this fault, the method involves memorizing, as a correction image, a destructured image of the scene to be observed; this destructured image is generated by calculating the means of the signals obtained when the selected scene is observed while the camera settings are simultaneously modified inconsistently; the correction image is then subtracted from the image formed by the signals output by the detector during real observation of the scene selected with correct camera settings. Application to all scanning cameras.