Method of and arrangement for image data compression by means of a neural network
First Claim
1. A method used in an arrangement for image data compression by vector quantization in accordance with a precoding in blocks constituted by a plurality of picture elements, thereafter comparing by means of a neural network precoded blocks with reference words stored in the form of a code book in the neural network so as to transmit to a receiver selected indices corresponding to the index of the reference words nearest to the precoded blocks, characterized in that it includes a stage of generating an adaptive code book by means of the neural network, the stage comprising the following steps:
- A--feeding the neural network with prototypes which prescribe the states of stable neurons in the neural network which determines its synaptic coefficients associated with each prototype during the learning phases,B--resolving by means of the neural network with the aid of test vectors supplied by the pictures by effecting in accordance with consecutive dynamic relaxation phases performed by feeding back the input to the output of the neural network, after which the states of the neurons are stabilized in configuration denoted attractors around which the test vectors are grouped in clouds,C--comparing each test vector with its corresponding attractor, with determination of the distance by which they are separated from each other and determination of the rate of frequentation of each attractor by the test vector,D--determining the rough mean of all the distances;
and when this rough mean exceeds a predetermined value, selecting attractors whose rate of frequentation exceeds a predetermined rate, thereafter determining the centre of gravity of the total cloud existing around each selected attractor to generate an optimum attractor when this attractor is the better and preserve the old attractor in the opposite case so as to form a group of new attractors which are then used again as new prototypes to restart the method of step A,and when this rough mean is less than the predetermined value, to utilize all the final attractors, to form the reference words of the adaptive code book which are stored in a synaptic coefficients store of the neural network, said reference words being transmitted to the receiver to update its local code book,the method also including a coding stage utilizing the said reference words to encode the rough image and thereafter to select the indices to be transmitted.
1 Assignment
0 Petitions
Accused Products
Abstract
Method of and arrangement for image data compression by vector quantization in accordance with a precoding in blocks, thereafter comparing by means of a neural network precoded blocks with reference words stored in the form of a code book so as to transmit selected indices to a receiver. In accordance with the method, the neural network effects a learning phase with prescribed prototypes, thereafter with the aid of test vectors originating from the image generates an adaptive code book which is transmitted to the receiver. This adaptation utilizes attractors, which may be induced metastable states, of the neural network, and which are submitted to an optimizing procedure. The arrangement can process images with a view to their storage. It is also possible to utilize two devices which operate alternately, one device for generating the adaptive code book and the other one to utilize it with the object of processing television pictures in real time.
31 Citations
10 Claims
-
1. A method used in an arrangement for image data compression by vector quantization in accordance with a precoding in blocks constituted by a plurality of picture elements, thereafter comparing by means of a neural network precoded blocks with reference words stored in the form of a code book in the neural network so as to transmit to a receiver selected indices corresponding to the index of the reference words nearest to the precoded blocks, characterized in that it includes a stage of generating an adaptive code book by means of the neural network, the stage comprising the following steps:
-
A--feeding the neural network with prototypes which prescribe the states of stable neurons in the neural network which determines its synaptic coefficients associated with each prototype during the learning phases, B--resolving by means of the neural network with the aid of test vectors supplied by the pictures by effecting in accordance with consecutive dynamic relaxation phases performed by feeding back the input to the output of the neural network, after which the states of the neurons are stabilized in configuration denoted attractors around which the test vectors are grouped in clouds, C--comparing each test vector with its corresponding attractor, with determination of the distance by which they are separated from each other and determination of the rate of frequentation of each attractor by the test vector, D--determining the rough mean of all the distances; and when this rough mean exceeds a predetermined value, selecting attractors whose rate of frequentation exceeds a predetermined rate, thereafter determining the centre of gravity of the total cloud existing around each selected attractor to generate an optimum attractor when this attractor is the better and preserve the old attractor in the opposite case so as to form a group of new attractors which are then used again as new prototypes to restart the method of step A, and when this rough mean is less than the predetermined value, to utilize all the final attractors, to form the reference words of the adaptive code book which are stored in a synaptic coefficients store of the neural network, said reference words being transmitted to the receiver to update its local code book, the method also including a coding stage utilizing the said reference words to encode the rough image and thereafter to select the indices to be transmitted. - View Dependent Claims (2, 3, 4, 7, 8, 9, 10)
-
-
5. An arrangement for image data compression by vector quantization in accordance with a precoding in blocks constituted by a plurality of picture elements, thereafter comparing by means of a neural network precoded blocks with reference words stored in the form of a code book in the neural network so as to supply selected indices corresponding to the indices of reference words which are closest to the precoded blocks, the neural network having a synaptic coefficients memory and a calculating member, characterized in that the arrangement comprises:
-
a neural network provided with learning means, a memory for storing test vectors coming from the image, a memory for storing prototypes prescribing stable states to the neural network, said two memories applying their data to the neural network which, for each prescribed prototype, determines its corresponding synaptic coefficients during learning phases, thereafter performs the resolving phases with the test vector by effecting in accordance with a relaxation dynamic at the end of which the states of the neurons stabilize in configuration denoted attractors around which the test vectors are regrouped into clouds, the neural network being fed back to itself, and an adaptive circuit which compares each test vector with its corresponding attractor, which determines the distance which separates them from each other, which determines the rate of frequentation of each attractor by the test vectors, which takes the rough mean of all the distances, (i) and when this rough mean exceeds a predetermined value, selects the attractors whose rate of frequentation exceeds a predetermined rate, thereafter determines the centre of gravity of the total cloud existing around each selected attractor to generate a recentred attractor when it is the better and to preserve the old one in the opposite case so as to form a group of new attractors which are used as new prescribed prototypes in the neural network to perform at least a new adaptation cycle, (ii) and when this rough mean of all the distances is less than the predetermined value, the arrangement uses all the final attractors to form the reference words of the adaptive code book; which are stored in the synaptic coefficients code book, which are used in the encoding of the overall image, which are stored in an attractor memory to be used by an indexing circuit to select the indices to be transmitted, which are transmitted to the receiver to update the local code book. - View Dependent Claims (6)
-
Specification