Pattern recognition system

Claim 1

A self-organizing pattern recognition system for identifying and categorising an input pattern having one or more identifiable elements comprising: a pattern generation unit (F1) for receiving the input pattern (12) and for generating an output pattern in response to the elements of the input pattern; adaptive filter means (14)for individually weighting elements of the output pattern relative to each of a plurality of pattern categories; a category selection unit (F2) for selecting at least one pattern category in response to combinations of weighted elements; template means (16) for generating and supplying to the pattern generation unit (F1) a template or expected pattern in response to the at least one selected pattern category; means for generating a determinant (18) related to the coincidence of elements between the input pattern and the expected pattern; and means for modifying the weighting of elements of the output pattern and the expected pattern relative to the at least one selected pattern category so that where the determinant exceeds a threshold value the means for modifying is operable to retain elements defining a modified expected pattern in common with the input pattern and the expected pattern and to deemphasise all other such elements and where the determinant fails to exceed the threshold value the means for modifying is operable to select an alternative pattern category so that a new expected pattern may be generated utilising the alternative pattern category.

Claim 4

A system as claimed in either of Claims 2 or 3, wherein the means for modifying includes means for reducing to approximately zero the weighting of elements relative to the at least one selected pattern category, which are not present in the coincident pattern.

Claim 5

A system as claimed in any one of Claims 2 to 4, wherein the coincident pattern is the input pattern when no expected pattern is generated and the coincident pattern is a pattern of elements common to the input pattern and the expected pattern when an expected pattern is generated.

Claim 7

A system as claimed in any one of Claims 2 to 6, wherein the threshold value is a predetermined ratio of the number of elements in the coincident pattern relative to the number of elements in the input pattern.

Claim 8

A system as claimed in any one of Claims 2 to 7, wherein the category selection unit (F2) selects a pattern category according to the ratio of (a) the product of a constant α and the number of elements in the coincident pattern to (b) the sum of a constant β and the number of elements in the expected pattern.

Claim 9

A system as claimed in any one of the preceding claims wherein the category selection unit (F2) selects only the single pattern category having the largest sum of weighted elements of the output pattern.

Claim 10

A system as claimed in any one of the preceding claims wherein the means for modifying alters the weighting of the elements of the ouput pattern in the adaptive filter means in inverse proportion to the number of elements common to the input pattern and the expected pattern.

Claim 12

A system as claimed in any one of the preceding claims wherein the initial weighting of all possible elements in the output pattern is low.

Claim 14

A system as claimed in any one of the preceding claims further comprising means (50) for changing the threshold value which results in modification of the adaptive filter means and the template means.

Claim 15

A system as claimed in any one of the preceding claims comprising means for selecting a further pattern category prior to receipt of a further input pattern by the system to anticipate an input pattern corresponding to the selected further pattern category dependant upon a previous input pattern.

Claim 16

A method of pattern recognition by identification and categorisation of an input pattern having one or more identifiable elements comprising: supplying the input pattern to a pattern generation unit (F1) which generates in response to the elements of the input pattern an output pattern; individually weighting elements (14) of the output pattern relative to each of a plurality of pattern categories; selecting at least one pattern category (F2) based on combinations of weighted elements; generating (16) and supplying to the pattern generation unit (F1) a template or expected pattern in response to the at least one selected pattern category; generating a determinant (18) related to the coincidence of elements between the input pattern and the expected pattern and detecting whether the determinant exceeds a threshold value; and when the determinant exceeds the threshold value, modifying the weighting of the elements of the output pattern and the elements of the expected pattern relative to the at least one selected pattern category to retain elements defining a modified expected pattern common with the input pattern and the expected pattern and to deemphasise all other such elements and, when the determinant fails to exceed the threshold value, selecting an alternative pattern category and utilising the alternative selected pattern category to generate a new expected pattern.

Claim 18

A method as claimed in either of Claims 16 or 17, wherein the determinant is generated by determining the number of elements common to the input pattern and the expected pattern relative to the number of elements in the input pattern.

Claim 19

A method as claimed in any one of Claims 16 to 18, wherein the weighting of elements of the output pattern is modified in inverse proportion to the number of elements common to the input pattern and the expected pattern.