Abstract: The disclosed method comprises a learning phase, during which the object zone to be recognized, selected in a learning image, is modelized, and a recognition phase during which the modelized zone is compared with image zones of a same size taken taken from the images to be processed.

Abstract: A device for computing 16 mono-dimensional direct cosine transforms from 16 luminance values having an input memory. A first computing device delivers respectively two values of the form A+B and (A-B).times.D, where A and B are two values applied respectively to two inputs of the first computing device and where D is a predetermined coefficient. A first intermediate value storage device uses the first computing device for 32 similar operations. A second computing device delivers a value of the form E+G, where E and G are two values applied consecutively to the input of the second computing device. A second intermediate value storage device reuses the second computing device for 17 additional operations. Finally, a device changes the order of the transformed values.

Abstract: This process uses a bidirectional transformation by blocks for images representing highly animated scenes and also uses a differential coding of the transformation coefficients when the pictures show slightly animated scenes. The coding device mainly comprises a cosine transformation device, a subtractor, and inter-image decoding device, a multiplexer, a device for discriminating the blocks showing highly animated scenes and blocks showing slightly animated scenes. The latter device controls the multiplexer. The subtractor and the decoding device make it possible to calculate the difference between each transformation coefficient and the homologous coefficient of the homologous block in the preceding picture. Moreover, a weighting device makes it possible to transmit with a larger number of bits the coefficients or coefficient differences corresponding to the low spatial frequencies of the images.