Abstract: A system (22) for synthesis of textured images, comprising a texture analyser (24) and a texture synthesizer (28). The texture analyser (24) comprises an analyzing neural network (30) which learns to characterize a texture by calculating synaptic coefficients (C.sub.ab), utilizing at least one proximity function which characterizes a neighbourhood around pixels of said texture. The texture synthesizer (28) comprises a synthesizing neural network (40) which receives the synaptic coefficients (C.sub.ab) thus calculated and which, utilizing a relaxation mechanism, synthesizes a replica of the texture learned. The neural networks (30), (40) may have a tree-type structure.

Abstract: A method is proposed for determining approximations of eigenvectors of a covariance matrix associated with signal data on the basis of the Principal Component Transform. Successive iterations on estimates enhance the development into the direction of the principal component considered. The method is applicable to the Singular Value Decomposition of data matrices. The operations to be performed are similar to those required in a neural net for adapting the synaptic coefficients, Apparatus are proposed for performing the method.

Abstract: Device for transmitting and/or storing signals presented thereto representing the textures of different areas of a video picture. The device includes a sampling stage for deriving a reference sample of the texture signals from a given picture area, and a coding stage for coding the reference sample signal as well as area signals representative of texture over the entire area from which the sample has been taken. The coded output signals of the coding stage may be transmitted by a transmission medium such as a communication channel and/or a store. A decoding stage then decodes the coded signals, and a texture synthesis stage uses the decoded sample signals to enhance the decoded area signals, thereby effectively achieving synthesis of a reproduction of the relevant picture area having a texture closely resembling the texture of such area of the original picture.

Abstract: A device for coding still images has a device for segmenting the images into homogeneous regions whereto a specific label is attached, and a device for coding the contents of the contour. The contour coding device has a sub-assembly which detects contours for each regions, for each region successively considered, it detects points which are to form a series of control points and transforming the part of the contour joining two successive control points into a straight segment. Then the validity criterion is verified. If the criterion is not met, at least one supplementary control point is introduced between the two control points and so on until it is. The transformation is applied until the contour loop is completely closed.

Abstract: Method and apparatus of error back-propagation for use in a neural network system. A first group (11) of processing devices (13.sub.1, 13.sub.2, 13.sub.3) performs the resolving steps and a second group (12) of analogous processing devices (13.sub.4, 13.sub.5) performs the training steps while backpropagating errors calculated in a central processing device (10). The synaptic coefficient matrix C.sub.ij of the first group and the transposed matrix T.sub.ji of the second group are simultaneously updated. This updating of the synaptic coefficients can be performed by means of multipliers (34.sub.1 to 34.sub.N) and adders (37.sub.1 to 37.sub.N).

Abstract: The invention relates to a norms computation procedure applied within a neuronal network structure or a computer. It permits determination of the norm of the new synaptic coefficients without necessitating their explicit computation. To this end, the structure comprises a processing unit 10 which determines output potentials y.sub.k and variation .DELTA..sub.k, a storage unit 13 for the old and new norms and a computational unit 15 for the new norms. The latter are thus rapidly determined without needing to call the synaptic coefficients in memory. The procedure can be utilized with certain algorithms which necessitate a computation of norms.

Abstract: Texture synthesis device and method, based on digital video signals. The synthesis device comprises a circuit (16) for computing and memorizing the structural characteristics, based on a sample referred to as reference sample, a circuit (17) for restituting an approximate synthesis picture by means of growth in a predetermined region, based on its characteristics and on the sample with a subdivision of said picture in blocks of an identical size, a circuit (18) for constituting a dictionary, based on the blocks of the sample of the same size as those of said picture and a circuit (19) for decorating said approximate synthesis picture, based on output signals of said circuit for constituting the dictionary. The invention also relates to a method of and a device for transmitting and/or storing texture signals, including the implementation of this method and this synthesis device.

Abstract: Method of processing signal data on the basis of Principal Component Transform, apparatus for performing the method. A method is proposed for determining approximations of eigenvectors of a covariance matrix associated with signal data on the basis of the Principal Component Transform. Successive iterations on estimates enhance the development into the direction of the principal component considered. The method is applicable to the Singular Value Decomposition of data matrices. The operations to be performed are similar to those required in a neural net for adapting the synaptic coefficients. Apparatus are proposed for performing the method.

Abstract: A neural integrated circuit, comprising a synaptic coefficient memory, a neuron state memory, resolving means and learning means which simultaneously operate in parallel on each of the synaptic coefficients in order to determine new synaptic coefficients. The learning means comprise means for performing a learning function on the states Vj of input neurons and on a correction element Si which is associated with each output neuron, and also comprise incrementation/decrementation elements which determine the new synaptic coefficients in parallel. The learning functions may be formed by logic AND-gates and exclusive-OR gates. The integrated circuit is used in a neural network system comprising a processing device.

Abstract: A data processing device for selecting data words which are contained in a dictionary and which are nearest to a data word to be processed according to a correspondence criterion. The device includes: first apparatus for segmenting the space enclosing the assembly of data words of the dictionary; second apparatus for generating, for each segment, sub-dictionaries by making an arbitrary segment correspond, in accordance with the correspondence criterion, to words of a sub-dictionary; third apparatus for utilising the sub-dictionaries by determining, for an arbitrary data word to be processed, the segment with which it is associated, followed by determination, in accordance with the correspondence criterion, of that word or words among the words of the sub-dictionary associated with the segment which corresponds (correspond) best to the arbitrary data word to be processed. Segmentation can be realised by means of a layered or tree-like neural network.

Abstract: Correlation device calculating a correlation between a current block of a first image and symbol blocks taken from a second image to determine the displacement vectors. Selection means (26) (27) select a limited number of symbols contained in the said blocks to which the correlation calculation relates. This selection can be effected by either a random choice or as a function of the data content of the blocks. The said current block can be divided into zones onto which the selection means act. A correlator (25) performs the correlation calculation block-sequentially on the overall current block and a decision unit (28) determines the displacement vector. It is possible to chose for the correlator (25) an architecture which promotes either its calculation rate or a reduction in its hardware bulk. The invention is used to reduce the flow of data to be transmitted, more specifically for image (television, videophone) or sound processing.

Abstract: A method structure for the compression of data utilizes an encoder which effects a transform with the aid of a coding neural network, and a decoder which includes a matched decoding neural network with effects almost the inverse transform of the encoder. The method puts in competition M coding neural networks (30.sub.1 to 30.sub.M) wherein M>1 positioned at the transmission end which effects a same type of transform and the encoded data of one of which are transmitted, after selection (32, 33) at a given instant, towards a matched decoding neural network which forms part of a set of several matched neural networks (60.sub.1 to 60.sub.Q) provided at the receiver end. Learning is effected on the basis of predetermined samples. The encoder may comprise, in addition to the coding neural network (30.sub.1 to 30.sub.M), a matched decoding neural network (35.sub.1 to 35.sub.M) so as to effect the selection (32, 33) of the best coding neural network in accordance with an error criterion.