Apparatus for recognizing alphanumeric characters

• US 6,363,171 B1
• Filed: 01/13/1995
• Issued: 03/26/2002
• Est. Priority Date: 01/13/1994
• Status: Expired due to Term

• Current Assignee: SGS-Thomson Microelectronics, S.R.L. (STMicroelectronics NV)
• Original Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Inventors: Kovacs, Zsolt M.
• Primary Examiner(s): Bella, Matthew C.
• Application Number:
  US08/373,937
• Time in Patent Office: 2,629 Days
• Field of Search: 382/156, 382/158, 382/161, 382/224, 395/21, 395/24, 706/15, 706/20, 706/26
• US Class Current: 382/161
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data

Abstract

An alphanumeric character image recognition system includes a first stage comprising at least a first, second and third digital image signal processing network having each at least one input terminal and at least one output terminal and said networks being designed to process image information from digital image signals, and comprising at least a first, second and third memory register having each at least one input terminal and at least one output terminal and the input terminals of the first, second and third memory registers being connected to the output terminal of the first network, the output of the second network and the output terminal of the third network respectively and said memory registers being designed to contain the image information processed by the first, second and third digital image signal processing networks, and a second stage characterized in that said second stage comprises at least one first and one second classifier network having each at least one first and one second input terminal and a plurality of output terminals, the first input terminal of the first classifier network being connected to the output terminal of the first memory register and the second input terminal of said first classifier network being connected to the output terminal of the second memory register and the first input terminal of the second classifier network being connected to the output terminal of the second memory register and the second input terminal of said second classifier network being connected to the output terminal of the third memory register and at least one third classifier network having a plurality of input terminals connected to a plurality of output terminals of the first and second classifier networks and a plurality of output terminals which are output terminals of the system and said classifier networks being designed to carry out consecutive statistical operations on the image information until there are obtained statistical values corresponding to a predetermined classification of images to be recognized.

Claims

1. A system for recognizing an alphanumeric character image, said system comprising:

a first stage including at least a first, second and third digital image signal processing networks having each at least one input terminal and at least one output terminal and said networks being designed to process image information from digital image signals and including at least a first, second and third memory registers having each at least one input terminal and at least one output terminal and the input terminals of the first, second and third memory registers being connected to the output terminal of the first network, the output terminal of the second network and the output terminal of the third network respectively and said memory registers being designed to contain the image information processed by the first, second and third digital image signal processing networks;

a second stage;

wherein the first network provides information related to a first feature vector of the image, the second network provides information related to a second feature vector of the image, the third network provides information related to a third feature vector of the image;

wherein said second stage includes at least one first and one second classifier network having each at least one first and one second input terminal and a plurality of output terminals, the first input terminal of the first classifier network being connected to the output terminal of the first memory register and the second input terminal of said first classifier network being connected to the output terminal of the second memory register and the first input terminal of the second classifier network being connected to the output terminal of the second memory register and the second input terminal of said second classifier network being connected to the output terminal of the third memory register;

at least one third classifier network having a plurality of input terminals connected to a plurality of output terminals of the first and second classifier networks and a plurality of output terminals which are output terminals of the system; and

wherein said classifier networks carry out consecutive statistical operations on said data from said first, second, and third processing networks until there are obtained statistical values corresponding to a predetermined classification of said image.

2. System in accordance with claim 1 characterized in that between an input of the system and the input terminal of at least one of said first, second and third signal processing networks is inserted at least one first filter having at least one input terminal and at least one output terminal and at least one first memory element having at least one input terminal connected to the output terminal of the first filter and at least one output terminal and said filter being designed to perform a normalization of digital signals of an image input and said memory element being designed to memorize digital signals of normalized images.

3. System in accordance with claim 1 characterized in that the first and second classifier networks are neural networks each including at least a first layer of neurons and each neuron of the first classifier network having at least one first and one second input terminal connected to the output terminal of the first memory register and to the output terminal of the second memory register respectively and at least one output terminal, and each of the neurons of the second classifier network having at least one first and one second input terminal connected to the output terminal of the second memory register and to the output terminal of the third memory register respectively and at least one output terminal.

4. System in accordance with claim 3 characterized in that the first classifier network and the second classifier network each include at least one second layer of neurons each having a plurality of input terminals connected to a plurality of output terminals of the first neuron layer.

5. System in accordance with claim 3 characterized in that the third classifier network is a neural network including at least one layer of neurons each having a plurality of input terminals connected to a plurality of output terminals of neurons of the first classifier network and neurons of the second classifier network and at least one output terminal which is an output terminal of the system.

6. System in accordance with claim 1 characterized in that the neurons of the first, second and third statistical networks are the sigmoidal type.

7. System in accordance with claim 1 characterized in that the second stage includes at least a fourth classifier network having at least a first and a second input terminal and the first input terminal of the fourth classifier network being connected to the output terminal of the first memory register and the second input terminal of said fourth classifier network being connected to the output terminal of the third memory register and a plurality of output terminals connected to a plurality of input terminals of the third classifier network.

8. System in accordance with claim 7 characterized in that the fourth classifier network is a neural network including at least one first layer of neurons and each of said neurons having at least one first and one second input terminal connected to the output terminal of the first memory register and to the output terminal of the third memory register respectively and at least one output terminal.

9. System in accordance with claim 8 characterized in that the fourth classifier network includes at least one second layer of neurons each having a plurality of input terminals connected to a plurality of output terminals of neurons of the first level.

10. System in accordance with claim 8 characterized in that the neurons of the fourth classifier network are the sigmoidal type.

11. System in accordance with claim 1 characterized in that the image information memorized in the first memory register is information on the position of the dark points of the image.

12. System in accordance with claim 1 characterized in that the image information memorized in the second memory register is information on the directions of the tangents to the points of said edge.

13. System in accordance with claim 1 characterized in that the image information memorized in the third memory register is information on the curvature of the edge of the image, said information expressed as contours at points of said edge.

14. Apparatus for recognizing an alphanumeric character from an image of said character, said apparatus comprising:

a first classifier network having a first input terminal coupled to receive a first feature vector of said character, a second input terminal coupled to receive a second feature vector of said character, and a plurality of output terminals;

a second classifier network having a first input terminal coupled to receive said second feature vector, a second input terminal coupled to receive a third feature vector of said character, and a plurality of output terminals;

a third classifier network having a plurality of input terminals coupled to said pluralities of output terminals of said first and second classifier networks, and a plurality of output terminals operable to carry a statistical value corresponding to a predetermined classification of said image; and

wherein said third classifier network is operable to carry out consecutive statistical operations on information received from said first and second classifier networks until said statistical value is generated.

15. The apparatus of claim 14 further comprising:

a first digital image signal processing network operable to generate said first feature vector from a first digital image signal;

a second digital image signal processing network operable to generate said second feature vector from a second digital image signal; and

a third digital image signal processing network operable to generate said third feature vector from a third digital image signal.

16. The apparatus of claim 15 further comprising:

a first memory in communication with said first processing and classifier networks and operable to store information carried by said first feature vector;

a second memory in communication with said second processing network and said first and second classifier networks, and operable to store information carried by said second feature vector; and

a third memory in communication with said third processing and said second classifier networks, and operable to store information carried by said third feature vector.

17. The apparatus of claim 15 further comprising:

a filter operable to normalize raw data representative of said image and to generate therefrom one or more of said first, second, and third digital image signals;

a memory in communication with one or more of said first, second, and third processing networks and operable to store information from those of said first, second, and third digital image signals generated by said filter.

18. The apparatus of claim 14 wherein said first and second classifier networks each comprise a neural network that includes a first layer of neurons,

each neuron of said first classifier network having first and second neural inputs respectively coupled to said first and second inputs of said first classifier network, and having at least one output terminal, and

each neuron of said second classifier network having first and second neural inputs respectively coupled to said first and second inputs of said second classifier network, and having at least one output terminal.

19. The apparatus of claim 18 wherein said first and second classifier networks each include a second layer of neurons each having a plurality of input terminals coupled to a plurality of said output terminals of said neurons of a respective one of said first layers.

20. The apparatus of claim 18 wherein said third classifier network comprises a layer of neurons each having a plurality of input terminals coupled to a plurality of said output terminals of said neurons of said first and second classifier networks, each of said third-network neurons having an output terminal coupled to one of said plurality of output terminals of said third classifier network.

21. The apparatus of claim 20 wherein said neurons of said first, second, and third classifier networks are sigmoidal neurons.

22. The apparatus of claim 14 further comprising a fourth classifier network having a first input terminal coupled to receive said first feature vector, a second input terminal coupled to receive said third feature vector, and a plurality of output terminals coupled to said input terminals of said third classifier network.

23. The apparatus of claim 14 wherein said first feature vector carries information including the positions of black pixels within said image.

24. The apparatus of claim 14 wherein said second feature vector carries information including the direction of the tangents at the points composing an edge of said image.

25. The apparatus of claim 14 wherein said third feature vector carries information including the contours formed by the points composing an edge of said image.

26. A method for identifying an alphanumeric character, comprising:

scanning said character to generate image data;

generating first feature data, second feature data, and third feature data from said image data;

processing said first and second feature data with a first neural network to generate a first signal;

processing said second and third feature data with a second neural network to generate a second signal; and

processing said first and second signals with a third neural network to identify said character.

27. The method of claim 26 wherein said image data includes light points having positions, and wherein said generating first feature data comprises providing said positions of said light points as said first feature data.

28. The method of claim 26 wherein said generating said second feature data comprises:

identifying one or more of the points that define an edge of said character;

calculating the first derivative of the curve formed by said edge at said identified points; and

providing said first derivative as said second feature data.

29. The method of claim 26 wherein said generating said third feature data comprises:

identifying one or more of the points that define an edge of said character;

calculating the second derivative of the curve formed by said edge at said identified points; and

providing said second derivative as said third feature data.

30. A method for identifying an alphanumeric character, comprising:

generating image data for said character;

generating first feature data, second feature data, and third feature data from said image data;

processing said first and second feature data with a first neural network to generate a first signal;

processing said second and third feature data with a second neural network to generate a second signal;

processing said first and third feature data with a third neural network to generate a third signal; and

processing said first, second, and third signals with a fourth neural network to identify said character.

31. The method of claim 30 wherein said first, second, and third neural networks each comprise multiple layers of intercoupled neurons.