Back to results
US 5,511,134 AGranted
Image recognition device and image recognition method
AssigneeMatsushita Electric Industrial Co., Ltd.
InventorsYasunori Kuratomi, Hisahito Ogawa
Life of this patent
Estimating the term…
Abstract
A two-dimensional bit image generated by the input device is processed by a four-layer neural network and recognized. The neurons in the second layer are connected only to the neurons of the first layer aligned in a specific direction, thereby enabling the second layer to extract line components in specific directions. The second layer is further divided into plural regions, and all neurons in each region are connected to one corresponding neuron in the third layer. The output of the third layer neurons thus express the position and degree of the extracted line component in the image. All of the neurons in the third layer are connected to all of the neurons in the fourth layer, and image recognition is possible by learning.
Claims14
An image recognition device for recognizing an input image by a neural network system, comprising:
- input means for inputting an image;
- a first layer having a plurality of neurons aligned in a plane for receiving said image on said first layer;
- a second layer having a line component sensitive section divided into subsections, each of which contains a plurality of line sensing neurons:
- (1) each sensitive to line components of said image extending in a single direction, and
- (2) comprising:
- (i) receiving means for receiving signals from at least three neurons in said first layer, said three neurons from which signals are received being aligned only in said single direction,
- (ii) multiplier means for multiplying predetermined non-negative synapse weights by the signals received from said at least three neurons, respectively, to produce at least three non-negative products,
- (iii) adder means for adding said products to produce a sum, and
- (iv) comparator means for comparing said sum with a predetermined threshold, and for producing no output signal when said sum is less than said predetermined threshold and producing an output signal relative to said sum when said sum is greater than said predetermined threshold;
- a third layer having a plurality of neurons, each sensitive to a group of neurons contained in said subsections;
- a fourth layer having a plurality of neurons, each sensitive to all said neurons in third layer;
- and
- a maximum detecting means for detecting a maximum from said neurons in said fourth layer.
10 more elements
An image recognition device for recognizing an input image by a neural network system, comprising:
- input means for inputting an image;
- a first layer having a plurality of neurons aligned in a plane for receiving said image on said first layer;
- a second layer having:
- (1) a first line component sensitive section divided into subsections each of which contains a plurality of first neurons, each sensitive to vertical line components of said image,
- (2) a second line component sensitive section divided into subsections each of which contains a plurality of second neurons, each sensitive to horizontal line components of said image,
- (3) a third line component sensitive section divided into subsections each of which contains a plurality of third neurons each sensitive to left-oblique line components of said image, and
- (4) a fourth line component sensitive section divided into subsections each of which contains a plurality of fourth neurons each sensitive to right-oblique line components of said image,
- each of said first, second, third and fourth neurons having:
- (i) receiving means for receiving signals from at least three neurons in said first layer, said three neurons from which signals are received being aligned in only a direction corresponding to the direction for which the neuron in the line component sensitive section of said second layer is assigned,
- (ii) multiplier means for multiplying predetermined non-negative synapse weights by the signals obtained from said at least three neurons, respectively, to produce at least three non-negative products,
- (iii) adder means for adding said products to produce a sum, and
- (iv) comparator means for comparing said sum with a predetermined threshold, and for producing no output signal when said sum is less than said predetermined threshold and producing an output signal relative to said sum when said sum is greater than said predetermined threshold;
- a third layer having a plurality of neurons, each sensitive to a group of neurons contained in said subsections;
- a fourth layer having a plurality of neurons, each sensitive to all the neurons in said third layer;
- and
- a maximum detecting means for detecting a maximum output from said neurons in said fourth layer.
13 more elements
An image recognition device for recognizing an input image by a neural network system, comprising:
- input means for inputting an image;
- a first layer having a plurality of neurons aligned in a plane for receiving said image on said first layer;
- a second layer having a geometric shape sensitive section divided into subsections, each of which contains a plurality of geometric sensing neurons:
- (1) each sensitive to a particular geometric shape in said image, and
- (2) comprising:
- (i) receiving means for receiving signals from at least three neurons in said first layer, said three neurons from which signals are received being aligned only in said particular geometric shape,
- (ii) multiplier means for multiplying predetermined non-negative synapse weights by the signals obtained from said at least three neurons, respectively, to produce at least three non-negative products,
- (iii) adder means for adding said products to produce a sum, and
- (iv) comparator means for comparing said sum with a predetermined threshold, and for producing no output signal said sum is less than said predetermined threshold and producing an output signal relative to said sum when said sum is greater than said predetermined threshold.
- a third layer having a plurality of neurons, each sensitive to a group of neurons contained in said subsections;
- a fourth layer having a plurality of neurons, each sensitive to all said neurons in said third layer;
- and
- a maximum detecting means for detecting a maximum output from said neurons in said fourth layer.
10 more elements
Dependencies are inferred from the claim text.