Method and apparatus for the classification of an article

Claim 3

A method as claimed in claim 1, wherein the components of the feature vector (X) are determined to meet said predetermined criterion if the closest target vector (W c ) has a predetermined relationship with the feature vector (X).

Claim 4

A method as claimed in claim 2, wherein the predetermined relationship is different depending on which target vector is closest.

Claim 5

A method as claimed in claim 1, wherein the components of the feature vector (X) are determined to meet said predetermined criterion if the individual components each meet a respective criterion.

Claim 6

A method as claimed in claim 4, wherein the components of the feature vector (X) are determined to meet said predetermined criterion if the difference between each individual component of target vector (W c ) and the corresponding component of the feature vector (X) is within a predetermined range.

Claim 7

A method as claimed in claim 1, wherein each of the target vectors associated with a target class lies within a respective Voronoi polygon, and wherein said predetermined criterion restricts the area of the Voronoi polygon within which a feature vector (X) is deemed to represent an article of the target class.

Claim 8

A method as claimed in claim 1, wherein the target vector (W c ) closest to the feature vector (X) is determined by measuring the Euclidean distance between the target vector (W c ) and the feature vector (X).

Claim 9

A method as claimed in claim 1, wherein at least one neural network is used to determine the target vector (W c ) closest to the feature vector (X).

Claim 10

A method as claimed in claim 8, wherein the neural network is an LVQ network.

Claim 11

A method as claimed in claim 8, wherein, for every target vector of a target class, the distance d to the feature vector (X) is calculated using one neuron or neuron-like part.

Claim 12

A method as claimed in claim 1, including the step of deriving the feature vector (X) using stored statistical data representative of the target class.

Claim 13

A method as claimed in claim 11, including the step of updating the statistical data representative of the target class on the basis of measurements of a test specimen determined to belong to that target class.

Claim 20

Apparatus for validating an article of currency, the apparatus comprising a measuring system, a preliminary processing system and a classification system for the classification of an article described by a k-dimensional feature vector (X) within at least n possible target classes, the preliminary processing system being responsive to measurements of physical features of a test item supplied by the measuring system for deriving the feature vector (X) and being operable to provide the feature vector (X) to the classification system, wherein the classification system comprises an output unit and a plurality of LVO neural network recognition units each connected at its input side to the preliminary processing system and its output side to the output unit, each recognition unit being operable to recognise exactly one class, the output unit being arranged to provide an output indicative of the determined class in response only to exactly one recognition unit providing an output indicative of the class.

Claim 21

Apparatus for validating an article of currency, the apparatus comprising a measuring system, a preliminary processing system and a classification system for the classification of an article described by a k-dimensional feature vector (X) within at least n possible target classes, the preliminary processing system being responsive to measurements of physical features of a test item supplied by the measuring system for deriving the feature vector (X) and being operable to provide the feature vector (X) to the classification system, wherein the classification system comprises an output unit and a plurality of LVO neural network recognition units each connected at its input side to the preliminary processing system and its output side to the output unit, each recognition unit being operable to recognise exactly one class, the output unit being arranged to provide an output indicative of the determined class in response only to exactly one recognition unit providing an output indicative of the class.

Claim 22

Apparatus for validating an article of currency, the apparatus comprising a measuring system, a preliminary processing system and a classification system for the classification of an article that can be described by k-dimensional feature vector (X), the preliminary processing system being responsive to measurements of physical features of a test specimen supplied by the measuring system for deriving the k-dimensional feature vector (X) and supplying the feature vector to the classification system, the classification system comprising a recognition unit for determining whether or not the article belongs to a target class representing a particular denomination in a particular orientation, the recognition unit being operable to determine which, amongst a plurality of target vectors associated with that target class, is the closest target vector (W c ) to the feature vector (X), and to designate the article as belonging to the target class if the components of the feature vector (X) meet a predetermined criterion indicating that the feature vector (X) lies within a predetermined boundary containing the closest target vector (W c ).

Claim 23

Apparatus for validating an article of currency, the apparatus comprising a measuring system, a preliminary processing system and a classification system for the classification of an article that can be described by k-dimensional feature vector (X), the preliminary processing system being responsive to measurements of physical features of a test specimen supplied by the measuring system for deriving the k-dimensional feature vector (X) and supplying the feature vector to the classification system, the classification system comprising a recognition unit for determining whether or not the article belongs to a target class representing a particular denomination in a particular orientation, the recognition unit being operable to determine which, amongst a plurality of target vectors associated with that target class, is the closest target vector (W c ) to the feature vector (X), and to designate the article as belonging to the target class if the components of the feature vector (X) meet a predetermined criterion indicating that the feature vector (X) lies within a predetermined boundary containing the closest target vector (W c ).

Claim 24

Apparatus according to claim 17, wherein the recognition unit comprises an input layer, and a neuron layer connected at its input side to the input layer via input weighting components and at its output side to an output layer, the input weighting components defining the target vectors for the recognition unit.

Claim 25

Apparatus as claimed in claim 18, wherein the weighting components each has values which have been determined during a training process.

Claim 26

Apparatus as claimed in claim 18, including means for varying the values of weighting components in accordance with measured physical features of a test specimen when that test specimen has been found to belong to the target class of the recognition unit including those weighting components, and when the weighting components define the target vector closest to the feature vector (X) for that specimen.

Claim 27

A device as claimed in claim 17, wherein the preliminary processing system is operable to derive different feature vectors (X) for respective target classes on the basis of different sets of statistical data each associated with a respective denomination, the respective feature vectors (X) each being applied to a respective recognition unit.

Claim 28

Apparatus as claimed in claim 21, including means for modifying the statistical data related to a denomination in accordance with measured physical features of an article which has been tested in response to determining that the article belongs to the class associated with that denomination.

Claim 29

Device comprising a measuring system, a preliminary processing system and a classification system for the classification of a pattern that can be described by a k-dimensional vector (X), especially a pattern of a banknote or a coin, within at least a number of possible target classes by means of the values of physical features supplied by the recording system, wherein the recording system, the preliminary processing system and the classification system are connected in the order in which they are listed substantially to form a chain at the output of which a service system is connected, characterised in that (a) the classification system comprises an output unit and several LVO neural network recognition units each connected at its input side to the preliminary processing system and at its output side to the output unit, and (b) a determined class of the pattern is transmitted by the output unit to an output channel if that class is recognised by a single recognition unit and no other recognition unit recognises a class.

Claim 30

Device comprising a measuring system, a preliminary processing system and a classification system for the classification of a pattern that can be described by a k-dimensional vector (X), especially a pattern of a banknote or a coin, within at least a number of possible target classes by means of the values of physical features supplied by the recording system, wherein the recording system, the preliminary processing system and the classification system are connected in the order in which they are listed substantially to form a chain at the output of which a service system is connected, characterised in that (a) the classification system comprises an output unit and several LVO neural network recognition units each connected at its input side to the preliminary processing system and at its output side to the output unit, and (b) a determined class of the pattern is transmitted by the output unit to an output channel if that class is recognised by a single recognition unit and no other recognition unit recognises a class.