Abstract: A device for encoding digital signals corresponding to images which are subdivided into blocks has a first encoding channel including a discrete cosine transform circuit. In addition to this first encoding channel the device has a second encoding channel which encodes the differences between the original coefficients present at the output of the discrete cosine transform circuit and the same coefficients after quantization and inverse quantization. This second channel encodes the differences after they are quantized with a finer quantization step than in the first encoding channel. The device further includes a prediction channel including a motion compensation stage, that uses both the inverse quantized coefficients coded in the first encoding channel and the inverse quantized coefficients coded in the second encoding channel to produce a predicted digital signal. This predicted digital signal is subtracted from the digital signal to be encoded by the device.