Apparatus and method for classifying and recognizing image patterns using neural network

• US 5,887,078 A
• Filed: 06/06/1996
• Issued: 03/23/1999
• Est. Priority Date: 12/29/1993
• Status: Expired due to Term

• Current Assignee: Korea Telecommunication Authority
• Original Assignee: Korea Telecommunication Authority
• Inventors: Kwon, Hee Yong, Kim, Dae Hwan, Kim, Byeong Cheol, Hwang, Hee Yeung, Cho, Dong Sub, Lee, Heung Ho
• Primary Examiner(s): Tran, Phuoc
• Application Number:
  US08/659,739
• Time in Patent Office: 1,020 Days
• Field of Search: 382/293, 382/295, 382/296, 382/298, 382/155, 382/124, 382/276, 382/156, 382/302
• US Class Current: 382/156
• CPC Class Codes: G06V 10/454 Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]

Abstract

The present invention provides an apparatus and a method for classifying and recognizing image patterns using a second-order neural network, thereby achieving high-rate parallel processing while lowering the complexity. The second-order neural network, which is made of adders and multipliers, corrects positional translations generated in a complex-log mapping unit to output the same result for the same object irrespective of the scale and/or rotation of the object. The present invention enables high-rate image pattern classification and recognition based on parallel processing, which is the advantage obtained in neural network models, because consistent neural networks and consistent network structure computation models are applied to all steps from the image input step to the pattern classifying and recognizing step.

Claims

1. An apparatus for classifying and recognizing an image received from an object, irrespective of scale and/or rotation of the image, the apparatus comprising:

a complex-log mapping means having optical input elements arranged in a polar exponential manner, for mapping the image received from the object into an orthogonal coordinate system, and for providing an output processed in parallel in the orthogonal coordinate system, the output being invariant to the scale and/or rotation of the image

a second-order neural network means for multiplying the output of the complex-log mapping means to produce a multiplication output and for adding the multiplication output to produce an output invariant to a positional translation caused by the complex-log mapping means; and

a classifying and recognizing neural network means, having a neural network, for classifying and recognizing the output of the second-order neural network means.

2. An apparatus in accordance with claim 1, wherein said complex-log mapping means comprises a plurality of optical input elements arranged in a plurality of concentric circles having exponentially increasing radii, are distributed in a polar exponential form, and have respective sampling regions having a predetermined phase difference arranged in alternating concentric circles;

and

wherein the concentric circles are arranged from a center of said complex-log mapping means to a peripheral edge of said complex-log mapping means, have an equal number of optical input elements, have equiangular positions of optical input elements, and have an equiangular distance between adjacent optical input elements.

3. An apparatus in accordance with claim 1, wherein the second-order neural network means comprises:

a first layer having a plurality of multipliers, each multiplier multiplying two values in the orthogonal coordinate system corresponding to two optical input elements and producing a multiplication output; and

a second layer having a plurality of adders coupled to the first layer, each adder adding multiplication outputs corresponding to two optical input elements spaced apart by a same vertical distance and a same horizontal distance in the orthogonal coordinate system.

4. An apparatus for classifying and recognizing an image received from an object, irrespective of scale and/or rotation of the image, the apparatus comprising:

a complex-log mapping means having optical input elements arranged in a polar exponential manner, for mapping the image received from the object into an orthogonal coordinate system, and for providing an output processed in parallel in the orthogonal coordinate system, the output being invariant to the scale and/or rotation of the image;

a second-order neural network means for ANDing and ORing the output of the complex-log mapping means, to produce an output invariant to a positional translation caused by the complex-log mapping means; and

a classifying and recognizing neural network means, having a neural network, for classifying and recognizing the output of the second-order neural network means.

5. An apparatus in accordance with claim 4, wherein said complex-log mapping means comprises a plurality of optical input elements arranged in a plurality of concentric circles having exponentially increasing radii, are distributed in a polar exponential form, and have respective sampling regions having a predetermined phase difference arranged in alternating concentric circles;

and

wherein the concentric circles are arranged from a center of said complex-log mapping means to a peripheral edge of said complex-log mapping means, have an equal number of optical input elements, have equiangular positions of optical input elements, and have an equiangular distance between adjacent optical input elements.

6. An apparatus in accordance with claim 4, wherein said second-order neural network means comprises:

a horizontal second-order neural network means for ANDing and ORing the output of the complex-log mapping means in the horizontal direction of the orthogonal coordinate system to produce an output; and

a vertical second-order neural network means for ANDing and ORing the output of the complex-log mapping means in the vertical direction of the orthogonal coordinate system to produce an output.

7. An apparatus in accordance with claim 4, wherein the second-order neural network means comprises:

a first layer having a plurality of elements, each element ANDing or ORing two values in the orthogonal coordinate system corresponding to two optical input elements and producing a first layer output; and

a second layer having a plurality of elements coupled to the first layer, each element ANDing or ORing two first layer outputs and producing a second layer output.

8. A method for classifying and recognizing an image received from an object through optical input elements disposed in a polar exponential arrangement, irrespective of scale and/or rotation of the image, the method comprising the steps of:

(a) mapping the image from the optical input elements to an orthogonal coordinate system to obtain an output irrespective of scale and/or rotation of the image;

(b) multiplying values of the output from step (a) to obtain multiplication outputs;

(c) adding the multiplication outputs from step (b) to obtain outputs irrespective of a position variation of the image; and

(d) classifying and recognizing outputs from step (c).

9. A method in accordance with claim 8, wherein the step of multiplying in step (b) comprises multiplying two values of the output from step (a) in the orthogonal coordinate system corresponding to two optical input elements and producing a multiplication output;

and

wherein the step of adding in step (c) comprises adding multiplication outputs from step (b) corresponding to two optical input elements spaced apart by a same vertical distance and a same horizontal distance in the orthogonal coordinate system.

10. A method for classifying and recognizing an image received from an object through optical input elements disposed in a polar exponential arrangement, irrespective of scale and/or rotation of the image, the method comprising the steps of:

(a) mapping the image from the optical input elements to an orthogonal coordinate system to obtain an output irrespective of scale and/or rotation of the image;

(b) horizontally and/or vertically ANDing and ORing values of the output from step (a) using a second-order neural network to produce an output from the output from step (a), irrespective of the position of the image; and

(c) classifying and recognizing the output from step (b).

11. A method in accordance with claim 10, wherein step (b) comprises:

processing the output from step (a) using a first layer of elements, each element of the first layer ANDing or ORing two values in the orthogonal coordinate system corresponding to two optical input elements and producing a first layer output; and

processing the first layer outputs using a second layer of elements, each element of the second layer ANDing or ORing two first layer outputs and producing a second layer output.