JPH06176158A - Pattern recognizing device - Google Patents

Abstract

PURPOSE:To realize highly accurate recognizing performance by respectively and individually recognizing plural kinds of feature vectors to an input pattern, and then integrating the results for final recognition. CONSTITUTION:Plural feature extracting parts 11 obtain respectively different feature vectors from the input pattern, plural single feature recognising parts 2 individually recognize them and a category recognizing part 13 recognizes the input pattern through the use of obtained similarity in this pattern recognizing device. In the single feature recognizing part 12, a fuzzy broadly-classifying part 22 obtains the group attribution degree of each category group to the input pattern and plural thinly-classifying parts 23 obtain the in-group similarity of each category within each category group so as to obtain the similarity of each category to the input pattern through the use of the group attribution degree and the in-group similarity. On the other hand, in the category recognizing part 13, an integrated similarity converting part 14 converts the similarity of each obtained category to the integrated similarity and a category discrimination part 15 obtains the final similarity of each category through the use of the integrated similarity so as to recognize.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device for recognizing an input pattern. In particular, the present invention relates to a device for recognizing a pattern by performing hierarchical recognition such as performing a major classification on an input pattern and then performing a fine classification.

[0002]

2. Description of the Related Art For a conventional input pattern,
First, a major classification is performed to select a category group to which the input pattern belongs (here, the category group refers to a set of patterns having similar pattern feature vectors), and then the subclassification is performed in the selected category group. As an example of a pattern recognition device for recognizing an input pattern by performing the above, for example, the IEICE Transactions D-II Vol.
75-D-II No.3 pp545-553 "Large-scale neural network" C
mbNET-II "".

FIG. 9 shows a block diagram of this conventional pattern recognition apparatus. Reference numeral 110 denotes a feature extraction unit for extracting a feature vector used to identify a category from an input pattern. 111 is a large classification unit, which roughly classifies the input pattern into each category group. 112
Is a sub-classification unit, which sub-classifies the input pattern within each category group. 113 is a group selection unit, which is a large classification unit 1
A plurality of category groups are selected from 11 output values (hereinafter referred to as “fitness”). Reference numeral 114 denotes a fine classification unit input signal selection unit, which inputs the feature vector of the input pattern based on the group selection information obtained by the group selection unit 113.
112 is to be selected. A discriminating unit 115 discriminates the input pattern from the conformance of the category group selected by the group selecting unit 113 and the output value of the fine classifying unit 112.

In the general classification unit 111, reference numeral 116 is an input unit for inputting a feature vector of an input pattern. Reference numeral 117 denotes a multi-input / single-output signal processing unit, which calculates the matching degree of each category group with respect to the input pattern.

In the fine classification unit 112, reference numeral 118 is an input unit for inputting the feature vector of the input pattern output from the fine classification unit input signal selection unit 114. Reference numeral 119 denotes a multi-input / single-output signal processing unit, which multiplies the output of the lower-layer input unit 118 or the multi-input / single output signal processing unit 119 connected to the multi-input / single-output signal processing unit and the weighting coefficient, which is the degree of connection thereof. The sum total is subjected to threshold processing and output. Here, these multiple multi-input one-output signal processing units have a layered structure and are networked so that signals are propagated only to the upper layer without mutual coupling in each layer, and thus within the category group for the input pattern. The degree of similarity to each category is required. Reference numeral 120 denotes a maximum value selection unit that selects the maximum value from the outputs of the plurality of multi-input / output signal processing units in the uppermost layer.

In the identification unit 115, 121 is a similarity calculation unit, and the similarity of each category is calculated from the goodness of fit of the category group selected by the group selection unit 113 and the output value of the subclassification unit 112 corresponding to the category group. It calculates the degree. A category identification unit 122 identifies the category of the input pattern by obtaining the maximum similarity of each category obtained from the similarity calculation unit 121.

The operation of the conventional pattern recognition apparatus configured as described above will be described below. First, the feature extraction unit 110 determines the feature vector X consisting of n feature data for the input pattern.

[0008]

[Equation 1]

Calculating Next, the feature vector X is input to the input unit 116 of the large classification unit 111. There are n input units 116, which are equal in number to the feature data of the pattern, and each feature data x _i is input to the corresponding input unit 116.
Each of the multi-input one-output signal processing units 117 of the large classification unit 111 has a weight coefficient v _ij (1 ≦ i ≦ m _r ; m _r that is a degree of connection of the input x _j of the input unit 111 connected thereto and After calculating the sum of the product of the number of category groups and 1 ≦ j ≦ n), this is calculated as a feature vector X and a weighting coefficient vector V _{i of} each multi-input one-output signal processing unit 117.

[0010]

[Equation 2]

Norm of | X |, | V_iDivided by the product of |
Output That is, the weighting coefficient vector shown in FIG.
Toru V_iOutput value sim of the multi-input single-output signal processing unit 117 having
(X, V _i) Is

[0012]

[Equation 3]

It can be expressed as It should be noted that the weighting coefficient vector V _i is designed in advance so that a predetermined multi-input / single-output signal processing unit generates a maximum output for a set of patterns having similar feature vectors X.

According to the conventional example, these weighting coefficient vectors V _i are designed by the following method. First, in the first process, every time the feature vector X of the pattern for designing the weighting coefficient vector is input, V _C having the largest sim (X, V _i ) is input.
The calculated (this time, X is that optimally matched to V _C.), Closer to V _C to X. Further, when the number of input patterns optimally matching one weighting coefficient vector exceeds a certain number, the area covered by the weighting coefficient vector is divided into two. In the second step, the optimum matching V _i is obtained for all the patterns for designing the weighting coefficient vector, and it is checked whether or not it has changed from the previous time. Then, if there is a change, the V _i is corrected. At this time, as in the first step, the weighting coefficient vector is also divided. This is repeated until the weight coefficient vector is corrected and there is no division.

By designing the weight coefficient vector in this manner, each weight coefficient vector V _i can be divided and quantized in the feature vector space of the input pattern. That is, the input pattern is classified into a set of a plurality of patterns having similar feature vectors, that is, a plurality of category groups by each weighting coefficient vector V _i . Then, the output value of each multi-input one-output signal processing unit 117 is used as the degree of conformity of each category group with respect to the input pattern, and the group selection unit 11
Output to 3.

The group selection unit 113 selects an arbitrary number of category groups in descending order of goodness of fit obtained by the large classification unit 111, and outputs group selection information indicating which category group has been selected and the goodness of fit corresponding thereto. To do.

On the basis of the group selection information obtained from the group selection unit 113, the subclassification input signal selection unit 114 selects the subclassification unit 112 to which the feature vector X of the input pattern is input and sets X to these subclassification units. Output to 112.

In each subclassification section 112 corresponding to the category group selected by the group selection section 113 (that is, the subclassification section 112 to which the feature vector X of the input pattern is input from the subclassification input signal selection section 114), First, the feature vector X is input to the input unit 118. The number of input units 118 is equal to the number of characteristic data items of the pattern, and each characteristic data item x _i is input to the corresponding input unit 118. Each multi-input one-output signal processing unit 119 of the sub-classification unit 112 multiplies the output of the lower-layer input unit 118 or the multi-input one-output signal processing unit 119 connected to the weighting coefficient which is the degree of the connection. After converting the sum of all things with a threshold function, the value is output to the upper layer. Here, the multi-input one-output signal processing unit 119 in the top layer of each subclassification unit 112 is set to the same number as the number of categories of patterns included in each category group, and each multi-input one output signal in the top layer is set. The processing unit 119 corresponds to each of these categories. The maximum value selection unit 120 selects the maximum output value among the output values of each multi-input one-output signal processing unit 119 in the uppermost layer, the category corresponding to this multi-input one-output signal processing unit 119, and its maximum output value. Is output.

The weighting factor of each multi-input / single-output signal processing unit 119 is such that for the feature vector X of the pattern having each category in the category group, the multi-input / single output of the highest layer corresponding to each category. The signal processing unit 119 is previously learned so as to generate the maximum output.

Specifically, such a weighting coefficient learning method is performed by a learning algorithm called an error back propagation method. For the error backpropagation method, for example, DE Ru
"Lea by melhart, GEHinton and RJ Williams
rning Representations by Back-Propagating Errors, "
Nature, vol.323, pp.533-536, Oct. 9, 1986.

The outline of the error back propagation method will be described below. First, the feature vector X of the pattern for weighting coefficient learning
Is input to the input unit 118 of the subclassification unit 112. As described above, each multi-input / single-output signal processing unit 119 has a lower-layer input unit 118 connected to it, or an output of the multi-input / single-output signal processing unit 119 and a weighting factor which is a degree of connection thereof. After the total sum of those multiplied by is converted by the threshold function, the value is output to the upper layer. Here, the error E between the output signal o _{k of} all the multi-input one-output signal processing units 119 of the uppermost layer and the desired output signal t _k (this is called a teacher signal) is obtained as in (Equation 4). .

[0022]

[Equation 4]

[Mathematical formula-see original document] Here, [Sigma] _p is a total sum regarding the number of patterns of the teacher signal. The purpose of learning is to determine the value of the weighting coefficient that minimizes the error E, and the change amount Δw _ij of the weighting coefficient between the multiple input / output signal processing units 119 is calculated based on ( _Equation 5). To be done.

[0024]

[Equation 5]

However, ε is a positive constant called a learning rate. The error E can be reduced by repeating such updating of the weighting coefficient based on (Equation 5) each time the feature vector of the learning pattern is input. When the error E becomes sufficiently small, the learning is terminated assuming that the output signal has become sufficiently close to the desired value.

According to such a weighting coefficient learning method,
For a pattern having each category in the category group, the highest input multi-input one-output signal processing unit 119 corresponding to each category can generate the maximum output.
Therefore, the plurality of input / output signal processing units 11
By selecting, among the nine, the one that produces the maximum output by the maximum value selection unit 120, it is possible to identify the category of the input pattern within each category group, that is, in each subclassification unit.

In the identification unit 115, first, in the similarity calculation unit 121, from the goodness of fit of the category group selected by the group selection unit 113 and the output value of the subclassification unit 112 corresponding to the category group (Equation 6). The similarity of each category obtained by the subclassification unit 112 is calculated using the formula, and the similarity is calculated by the category identification unit.
Output to 122.

[0028]

[Equation 6]

However, a and b are real constants. Finally, the category identifying unit 122 compares the similarities of the categories obtained from the similarity calculating unit 121 and outputs the category corresponding to the maximum similarity among them as the identification result.

[0030]

In pattern recognition, features that are effective in identifying patterns are extracted as feature vectors, but it is generally difficult to realize sufficient recognition ability with only a single feature vector. A high identification function can be realized by using a plurality of types of feature vectors. In other words, by using multiple types of features, even if a pattern that can only be ambiguous or misidentified with a single feature vector, accurate identification may be possible with different feature vectors. This is because there is a possibility that the identification performance can be improved by placing importance on.

However, in a pattern recognition device for hierarchically recognizing patterns having a large number of categories,
In a configuration in which only one feature extraction unit and a recognition unit are provided as in the conventional example, when using a plurality of types of feature vectors, it is necessary to collectively input these and perform recognition. Absent. In this case, compared to the case of using only a single feature vector, it is possible to achieve a certain degree of high discrimination performance, but the advantages obtained by using different feature vectors as described above cannot be fully utilized effectively. There is a problem. That is, when a plurality of types of feature vectors are used together, the identification performance can be improved, but using the recognition result,
It is very difficult to correctly recognize a pattern that is erroneously recognized with a certain feature vector as described above, based on the identification result of a different feature vector.

Further, when a plurality of types of feature vectors are used together, there is a problem in that recognition requires a lot of calculation time as the number of dimensions of the feature vector increases.

In view of the above problems of the conventional pattern recognition apparatus, the present invention recognizes with high accuracy by effectively utilizing the advantage of using plural kinds of feature vectors together when using plural kinds of feature vectors in the recognition apparatus. It is an object of the present invention to provide a pattern recognition device capable of realizing the above and further shortening the time required for recognition.

[0034]

SUMMARY OF THE INVENTION According to the present invention, a plurality of feature extraction units for obtaining different feature vectors from input patterns and a similarity degree, which is the degree to which the input pattern belongs to each category, are obtained from each of the feature vectors. A plurality of single feature recognition units to be obtained, and a category identification unit that identifies the input pattern by using the similarity obtained from the plurality of single feature recognition units, wherein the single feature recognition unit is A group dictionary that stores a plurality of group reference feature vectors that represent a category group composed of a set of patterns having similar feature vectors, and an input pattern that uses the group reference feature vector and the feature vector of the input pattern A fuzzy major classification unit that calculates a group membership degree, which is a degree of belonging to each category group, and the input pattern using the feature vector of the input pattern It is obtained from a plurality of sub-classification units that obtain the degree of similarity within a group, which is the degree of belonging to each category included in the category, a group selection unit that selects a plurality of category groups from the group membership degree, and the group selection unit. A subclassification section input signal selection section for selecting a subclassification section for inputting a feature vector of the input pattern based on group selection information, a group membership degree of the category group selected by the group selection section, and the subclassification section. A single feature similarity calculation unit that obtains the similarity of each category to the input pattern using the obtained in-group similarity is provided, and the single feature similarity calculation unit is the category group selected by the group selection unit. A plurality of multipliers for multiplying the degree of group membership and all the in-group similarities obtained from the subclassification section, which are input with the feature vector of the input pattern from the subclassification section input signal selection section, and for each category To the multiplier A category similarity calculation unit that selects a plurality of output values and calculates the sum of these output values is provided, and the category identification unit is obtained from the single feature recognition unit corresponding to each of the feature vectors. A plurality of integrated similarity conversion units for converting the similarity of each category into the integrated similarity by normalizing the similarity with the maximum value among the similarities, and the integrated similarity obtained from the plurality of integrated similarity conversion units. And a category determination unit that determines the final similarity of each category by using the degree and finally identifies the input pattern.

[0035]

According to the present invention, with the above-described structure, each feature extraction unit first extracts a different feature vector from the input pattern. Next, these feature vectors are input to the corresponding single feature recognition units.

In each single feature recognition section, first, the feature vector is input to the fuzzy major classification section. The fuzzy major classification unit reads all the group-referenced feature vectors stored in the group dictionary, and performs major classification on the feature vector by ambiguously defining the boundaries of each category group in the feature vector space. , The degree of group membership, which is the degree to which the input pattern belongs to each category group, is obtained. Using these group membership degrees, the group selection unit selects a plurality of category groups and outputs group selection information to the subclassification unit input signal selection unit and the corresponding group membership degree to the single feature similarity calculation unit. To do. The subclassification unit input signal selection unit outputs the feature vector to the subclassification unit corresponding to the group selection information, and each subclassification unit uses the feature vector to include the input pattern in each category group. The in-group similarity, which is the degree of similarity to each category, is calculated and output to the single feature similarity calculation unit.

The single feature similarity calculation unit is obtained from the subclassification unit to which the group membership of the category group selected by the group selection unit and the feature vector from the subclassification unit input signal selection unit are input. All the in-group similarities are multiplied by the multiplier and output to the category similarity calculation unit. In the category similarity calculation unit, a plurality of output values of the multiplier are selected for each category, and the sum of these output values is calculated.
This is the similarity of each category to the input pattern. That is, in each single feature recognition unit, this similarity is output to the category identification unit as the similarity obtained from each feature vector for the input pattern.

Finally, in the category identification section, first, each integrated similarity conversion section determines the similarity of each category obtained from the single feature recognition section corresponding to each of the feature vectors among the similarities. It is converted to an integrated similarity by normalizing with the maximum value. Next, the category determination unit obtains the final similarity of each category with respect to the input pattern by using these integrated similarities, and identifies the final input pattern.

As described above, the present pattern recognition apparatus performs the classification processing on a plurality of types of feature vectors by the separate single feature recognition units to obtain the similarity of each category to each feature vector, and then the category recognition unit. Uses all of these similarities to integrate the identification results obtained from each single feature recognition unit to recognize the final input pattern. Therefore, even a pattern that is erroneously recognized by a certain feature vector can be correctly recognized by the identification result by a different feature vector, and highly accurate recognition can be realized. Further, in the conventional method, the time required for recognition is increased by using a plurality of types of feature vectors, but in the present pattern recognition device, since each single feature recognition unit performs identification processing in parallel, the time required for recognition is increased. Compared to the conventional example,
Can be shortened.

[0040]

Embodiments of the present invention will be described below with reference to the drawings.

Generally, pattern data input to a pattern recognition device includes time-series patterns such as voices, characters,
Although there are spatial patterns such as images, etc., any pattern data may be used in the present invention.

FIG. 1A shows a block diagram of a pattern recognition apparatus according to an embodiment of the present invention. Figure 1
In (a), 11 is a feature extraction unit that extracts a feature vector used to identify a category from an input pattern. However, each of the plurality of feature extraction units extracts a different feature vector. Reference numeral 12 denotes a single feature recognition unit, which is prepared in the same number as that of the feature extraction unit 11, and calculates the degree of similarity of the input pattern belonging to each category from the corresponding feature vector. A category identification unit 13 identifies the input pattern by using the similarity of each category to each feature vector of the input pattern obtained from each single feature recognition unit 12.

In the category discriminating unit 13, 14 is an integrated similarity converting unit, which is prepared in the same number as the feature extracting unit 11 and the single feature recognizing unit 12, and each category obtained from the corresponding single feature recognizing unit 12 The similarity is converted into an integrated similarity by normalizing the similarity with the maximum value among the similarities. A category determination unit 15 obtains the final similarity of each category with respect to the input pattern using all the integrated similarities, and identifies the final category.

Further, the category judgment unit 15 shown in FIG.
Is a first embodiment of the category determination unit 15. Figure 1
In (a), 16 is an adder, which obtains the final similarity of each category by adding the integrated similarity of each category obtained from the plurality of integrated similarity calculation units 14 for each same category. . Reference numeral 17 denotes a maximum value selection unit, which selects the maximum value from the final similarity of each category.

FIG. 1B is a block diagram specifically showing the structure of the single feature recognition unit 12. In FIG. 1 (b),
Reference numeral 21 denotes a group dictionary, which stores a plurality of group reference characteristic vectors representing a category group composed of a set of patterns having similar characteristic vectors. Reference numeral 22 denotes a fuzzy major classification unit, which uses the group reference feature vector stored in the group dictionary 21 to calculate the degree of group membership, which is the degree to which the input pattern belongs to each category group. A sub-classification unit 23 calculates the intra-group similarity, which is the degree to which the input pattern belongs to each category included in the category group. A group selection unit 24 selects a plurality of category groups from the degree of group membership. Reference numeral 25 denotes a fine classification unit input signal selection unit, which inputs the feature vector of the input pattern based on the group selection information obtained from the group selection unit 24.
Is to be selected. Reference numeral 26 is a single feature similarity calculation unit, which obtains the similarity of each category to the input pattern from the group membership of the category group selected by the group selection unit 24 and the in-group similarity obtained by the subclassification unit 23. It is a thing.

In the single feature similarity calculation unit 26, 27 is a multiplier, which calculates the group membership degree of the category group selected by the group selection unit 24 and the feature vector of the input pattern from the subclassification unit input signal selection unit 25. This is for multiplying the input intra-group similarity obtained from the sub-classification unit 23. Reference numeral 28 denotes a category similarity calculation unit, which selects a plurality of output values of the multiplier 27 for each category and calculates the sum of these output values.

FIG. 2 is a block diagram specifically showing the configuration of the fuzzy large classification unit 22. In FIG. 2, reference numeral 31 denotes an input unit for inputting the feature vector of the input pattern. A distance calculator 32 calculates the distances between all the group reference feature vectors in the group dictionary 21 and the feature vectors. Reference numeral 33 is a divider, which calculates the reciprocal of the output of the distance calculation unit 32. Reference numeral 34 is an adder, which calculates the total sum of the outputs of the respective dividers 33. A multiplier 35 multiplies the output of the adder 34 and the output of the distance calculator 32. A divider 36 calculates the reciprocal of the output of the multiplier 35.

FIG. 3 is a block diagram specifically showing the configuration of the first embodiment of the subclassification section 23. In FIG. 3, reference numeral 41 denotes an input unit for inputting the feature vector of the input pattern output from the subclassification unit input signal selection unit 25.
Reference numeral 42 is a category dictionary, which stores a plurality of category reference feature vectors indicating representative values of each category of the pattern. Reference numeral 43 denotes a distance calculation unit that calculates the distances between all the category reference feature vectors of the category dictionary 42 and the feature vector of the input pattern. Reference numeral 44 is a divider, which calculates the reciprocal of the output of the distance calculation unit 43. Reference numeral 45 is an adder, which calculates the total sum of the outputs of the respective dividers 44. 46 is a multiplier, and an adder 45
Is multiplied by the output of the distance calculation unit 43. 47
Is a divider that calculates the reciprocal of the output of the multiplier 36.

The operation of the pattern recognition apparatus configured as described above will be described below.

First, different feature vectors X _k (k = 1 to N _F : N _F is the number of types of feature vectors) are extracted from the input patterns by the plurality of prepared feature extraction units 11. To do. Here, each feature vector X _k consists of n _k pieces of feature data and is expressed as in (Equation 7).

[0051]

[Equation 7]

As an example of a plurality of different feature vectors, if the pattern is a character, for example, the following may be considered. For example, if the character image information is 8
It is divided into 8 parts, and 64-dimensional grayscale features (mesh features) that have the frequency of black pixels in each region as an element and image information of characters are divided into 4 regions.
There is a 64-dimensional contour direction density feature that has the appearance frequency of the directional elements (4 directions of horizontal, vertical, right diagonal, and left diagonal) of the character contour points in each of the 4 divided areas.

Each feature vector X _k of the input pattern extracted by each feature extraction unit 11 corresponds to the corresponding single feature recognition unit 1
Entered in 2. Multiple single feature recognition unit 12
Respectively perform the following identification processing independently,
The similarity of each category with respect to each feature vector X _{k is obtained} . Note that the feature vector X _k used in the following description of the operation of the single feature recognition unit 12 is a feature vector obtained by a certain feature extraction method, and hereinafter, X _k and n _k are described as X and n for simplicity. To do.

The specific operation of the single feature recognition unit 12 will be described below. First, in the single feature recognition unit 12, the feature vector X is input to the fuzzy large classification unit 22. In the fuzzy major classification unit 22, the input unit 31 inputs the feature vector X and outputs X to the m _r distance calculation units 32. Each distance calculator 32
Is a group reference feature vector V _i (1 ≦ i ≦ m _r ; m _r is the number of group reference feature vectors, that is, the number of category groups) representing each category group stored in the group dictionary 21. Then, the distance d _i between X and V _i shown in (Equation 8) is calculated and output to the corresponding divider 33 and multiplier 35, respectively.

[0055]

[Equation 8]

However, f is a real number satisfying f> 1. Each divider 33 calculates the reciprocal of the distance d _i and outputs the output to the adder 34. The adder 34 calculates the sum of the outputs of all the dividers 33 and outputs the output to the m _r multipliers 35. Each of the multipliers 35 multiplies the output of the corresponding distance calculation unit 32 and the output of the adder 34, and inputs the output to the corresponding divider 36. Each divider 36 calculates the reciprocal of the output of the corresponding multiplier 35. Finally, in the fuzzy major classification unit 22, the output of each divider 36 is the group membership degree μ _i (1 ≦ i ≦ of each category group with respect to the feature vector X of the input pattern).
m _r ) is output to the group selection unit 24. That is, the group membership degree μ _i (1 ≦ i ≦ m _r ) of each category group is

[0057]

[Equation 9]

It can be expressed as The group reference feature vector representing each category group stored in the group dictionary 21 is a conventional clustering method, for example,
The Institute of Electronics, Information and Communication Engineers Makoto Nagao "Pattern Information Processing"
The K-means algorithm shown in (Corona), and
Isodata algorithm, Y.Linde, A.Buzo, and RMGr
"An Algorithm for Vector Quantizer desig" by ay
n, "IEEE Trans. Commun., COM-28, 1, pp.84-95, Jan.1
It is designed using the LBG algorithm shown in 980.

In the following, the group dictionary is calculated using the K-means algorithm.
A brief description of how to design 21. (1) m _r pieces (where m _r is a predetermined number of category groups) from a set of patterns for designing a group dictionary of recognition objects
, And select these feature vectors as m _r
It is assumed that each group reference feature vector V _i (1 ≦ i ≦ m _r ). (2) The distance d _i shown in (Equation 10) for each of the feature vectors X of all the group dictionary design patterns.

[0060]

[Equation 10]

V _i that minimizes is obtained. At this time, X
Shall belong to the category group S _i (1 ≦ i ≦ m _r ). (3) The average value of the feature vector X of the pattern belonging to each S _i is obtained, and this is set as V _i ′. (4) If V _i ′ = V _i holds for all i, the group reference feature vector V _i at this time is stored in the group dictionary 21. If not, V _i 'is set as a new group reference pattern signal V _i and the process returns to (2).

By designing the group reference feature vector in this manner, all patterns can be divided into some subsets (category groups) of patterns having similar feature vectors. The Isodata algorithm and LBG algorithm are basically the same as the K-means algorithm.

The group selection unit 24 selects a plurality of category groups in the descending order of the degree of group membership obtained by the fuzzy large classification unit 12, and selects group selection information indicating which category group has been selected by the fine classification input signal selection. It outputs the group membership degree corresponding to the unit 25 to the single feature similarity calculation unit 26. As a method of selecting a category group, a category group having a group membership degree equal to or higher than a certain threshold may be selected.

On the basis of the group selection information obtained from the group selection unit 24, the subclassification input signal selection unit 25 selects the subclassification unit 23 to which the feature vector X of the input pattern is input, and sets the feature vector X to these subclassifications. Output to the classification unit 23.

Each subclassification section 23 (that is, the subclassification input signal selection section 25) corresponding to the category group selected by the group selection section 24.
Subclassification section that receives the feature vector of the input pattern from
In 23), first, the input unit 41 inputs the feature vector X of the input pattern and outputs X to the m _c distance calculation units 43.
Each distance calculation unit 43 calculates a category reference feature vector W _i (1 ≦ i ≦ m _c ; m _c is the number of category reference feature vectors) representing the representative value of each category stored in the category dictionary 42. The distance d _i between X and W _i shown in (Equation 11) is read out, and the corresponding divider 44 and multiplier 46 are respectively calculated.
Output to.

[0066]

[Equation 11]

However, f is a real number satisfying f> 1. Each divider 44 calculates the reciprocal of the distance d _i and outputs its output to the adder 45. The adder 45 calculates the total sum of the outputs of all the dividers 44 and outputs the output to the m _c multipliers 46. Each of the multipliers 46 multiplies the outputs of the corresponding distance calculation unit 43 and the adder 45, and inputs the output to the corresponding divider 47. Each divider 47 calculates the reciprocal of the output of the corresponding multiplier 46. Finally, in the sub-classification unit 23, the output of each divider 47 is the single feature similarity ν _i (1 ≦ i ≦ m _c ) of each category with respect to the feature vector X of the input pattern. It is output to the calculation unit 26. That is, the in-group similarity ν _i (1 ≦
i ≦ m _c ) is

[0068]

[Equation 12]

It can be expressed as The category reference feature vector indicating the representative value of each category stored in the category dictionary 42 is an output of the distance calculation unit 43 corresponding to each category for an input pattern having each category in the category group. Is designed in advance so as to generate the minimum output as compared with the outputs of the other distance calculation units 43.

The method for designing the feature vector for category reference is performed by, for example, a learning algorithm called a learning vector quantization method (LVQ). For the learning vector quantization method, see, for example, "Le by T. Kohonen.
arning Vector Quantization for Pattern Recognitio
n ", Helsinki University of Technology, Report TKK-
It is shown in F-A601 (1986.11).

The learning vector quantization method will be briefly described below. First, a category reference feature vector W _i having m _c category labels is prepared. As an initial value of this W _i , a feature vector of a pattern arbitrarily selected for each category from a pattern for designing a category dictionary composed of a pattern set included in each category group and a group dictionary 11 at the time of designing A reference feature vector obtained from a conventional clustering method such as the K-means algorithm described above is used. Then, the selected feature vector X of the pattern with any one category C _X from the pattern for the category dictionary design, this X sequentially, the following steps are repeated. (1) Select the category reference feature vector W _C closest to X. However, the category label of this W _C is C _C. (2) If C _X = C _C ,
W _C approaches the X direction. On the other hand, if C _X ≠ C _C , W _C is moved away from X. In addition, category reference feature vectors other than W _C are not updated.

The above-mentioned updating of the category reference feature vector when X is presented is repeated for all the prepared categories dictionary design patterns.

By designing the category reference feature vector of the category dictionary 32 in this way, for the feature vector of the pattern having each category in the category group, the category reference feature vector having the label of each category is always used. The feature vectors will be located at the closest distance. Therefore, the category of the input pattern can be recognized in each category group by selecting the distance calculation section 43 that generates the minimum output among all the distance calculation sections 43.

In the single feature similarity calculation unit 26, the multiplier 27 first causes the group membership degree of the category group selected by the group selection unit 24 and each subclassification unit 23 (that is, each subclassification unit 23 corresponding to the category group). , All the in-group similarities obtained from the fine classification unit 23) to which the feature vector of the input pattern is input from the fine classification input signal selection unit 25 are multiplied, and the outputs are output to the category similarity calculation unit 28. That is, the number of multipliers 17 provided is (the number of group-reference feature vectors × the sum of the number of category-reference feature vectors in each subclassification unit), and one of the category groups p selected by the group selection unit 24 is selected. Group membership degree μ _p (1 ≦ p ≦
m _r ; m _r is the number of category groups) and the in-group similarity ν _pq (1 ≦ q ≦ m _c ; m _C is pattern data for a certain category q obtained from the subclassification unit 13 corresponding to the category group p Number of categories) is input. That is, the output value ξ _pq of the multiplier 17 is

[0075]

[Equation 13]

It is represented by The category similarity calculation unit 28 classifies the output values of all the multipliers 27 into each category, and selects a plurality of output values having a large output value.
Then, the sum of these selected output values is calculated for each category, and the sum of the similarity r _i (1 ≦ i ≦ N _C ; N _C of the categories to the input pattern in the single recognition unit 12 is the category Number) to the category identification unit 13. As a method of selecting a plurality of output values of the multiplier 27 for each category, it is possible to select an output value of the multiplier 27 having a certain threshold value or more.

As described above, the single feature recognition unit 12
For each of the input patterns for a certain feature, the feature vector obtained from the certain feature extraction unit 11 is subjected to a large classification, a fine classification, and finally a hierarchical identification that integrates the outputs thereof. Find the similarity of categories.

In the category identifying section 13, the similarity of each category with respect to each feature vector of the input pattern obtained from each single feature recognizing section 12 is first input to the corresponding integrated similarity converting section 14, and each category is input. The degree of similarity of is converted into an integrated degree of similarity. Next, the category determination unit 15 performs final recognition of the input pattern using the integrated similarities of all the categories obtained from the integrated similarity conversion units 14. That is, the category determination unit 15 performs the final identification of the classification result obtained by performing the recognition process of each feature vector of the input pattern in each single feature recognition unit 12, that is, the category identification. Is. In addition, the integrated similarity conversion unit
The category determination unit 15 converts the similarity obtained from the single feature recognition unit 12 into an integrated similarity that allows the identification results regarding each feature vector to be integrated well.

Specific operations of the integrated similarity conversion unit 14 and the category determination unit 15 will be described below.

In the integrated similarity conversion unit 14, as shown in (Equation 14), the similarity r _ki (1 ≦ k ≦ N _F : N _F of each category obtained from the corresponding single feature recognition unit 12 is a feature. Normalization is performed by dividing (the number of vector types, 1 ≦ i ≦ N _C ; N _C is the number of categories) by the maximum value r _{km ax} among these similarities, and the integrated similarity t _ki ( However, 0 ≦ t _ki ≦
1, 1 ≦ i ≦ N _C ; N _C is the number of categories), and output to the category determination unit 15.

[0081]

[Equation 14]

In the category determination unit 15, the adder 17 adds the integrated similarity t _ki of each category obtained from each integrated similarity conversion unit 14 for each same category as shown in ( _Equation 15). , The final similarity u _i (where 1 ≦ i ≦ N _C ; N _C is the number of categories). Finally, the maximum value selection unit 18 compares the final similarities u _i and outputs the category corresponding to the maximum similarity among them as the final recognition result.

[0083]

[Equation 15]

By the way, the most important value for the category discriminating unit 15 in the category discriminating unit 13 to make the final discrimination is the maximum value of the similarity of each category obtained from each single feature recognition unit 12 and its vicinity. , That is, the value of the similarity of the first, second, ... Candidates obtained by each single feature recognition unit 12.
However, the maximum similarity obtained from each single feature recognition unit 12 is affected by the total number of output units of each subclassification unit in the single feature recognition unit 12 (specifically, the output units of the subclassification unit). The larger the total number is, the smaller the maximum similarity tends to be.) Therefore, if the similarity obtained from each single feature recognition unit 12 is directly output to the category determination unit 15, the maximum similarity is always averaged. Single feature recognition unit with high degree
The identification result from 12 has priority. Therefore, as it is, it is difficult to recognize the advantage of using a plurality of types of feature vectors, that is, to correctly recognize a pattern that is erroneously recognized by a certain feature vector based on the identification result by different feature vectors.

However, in the integrated similarity conversion unit 14, by normalizing the similarity of each category obtained from each single feature recognition unit 12 by its maximum similarity, the maximum similarity of each single feature recognition unit 12 is increased. The similarity is always 1. Therefore, in the final category identification in the category determination unit 15,
The identification result of each single feature recognition unit 12 can be equally contributed, and the identification result from a certain single feature recognition unit 12 is not always given priority.

Here, the results obtained experimentally are shown. The pattern data to be recognized was a character of multi-font (Mincho, Gothic, etc.) 3390 character type (symbol, alphanumeric, hiragana, katakana, Greek character, JIS first level kanji). The feature vector for feature extraction is a 64-dimensional contour direction density feature and a 32-dimensional background density feature. The contour direction density is a feature amount indicating the direction and complexity of the character stroke in the local area, and the background density feature is a feature amount indicating the complexity of the background of the character. The detailed feature extraction method is Waki, Fujiwara, Takenouchi, Yokoe, and Shimizu, "Document fair copy system (3) -character recognition algorithm and its hardware-", 1986 IEICE General Conference, 1512,
6-154.

Single feature recognition unit corresponding to each feature vector
Group dictionary 21 in 12 and category dictionary of each subclassification unit 23
42 was designed with 27 fonts of 8 fonts. here,
The fuzzy major classification unit 22 in each single feature recognition unit 12 sets the number of category groups (the number of group reference feature vectors) to 7.
Further, the number of category reference feature vectors of each subclassification unit 23 is 10 in the category dictionary 42 corresponding to the contour direction density feature.
The total number of 70,949,957,562,1029,920,996 is 6483, and 112 in the category dictionary 42 corresponding to the background density feature.
The total number of 2,902,633,1287,751,1090,984 was 6769.

At the time of recognition, f in the fuzzy major classification unit 22 (Equation 8) in each single feature recognition unit was set to 1.2, and f in the fine classification unit 23 (Equation 11) was set to 1.3. The number of category groups selected by the group selection unit 24 is 7, that is, each feature vector is input to all the subclassification units 23. Further, in the category similarity calculation unit 28, the number of output values of the multiplier 27 selected for each category is set to 4.

For comparison, a single feature recognition section 12 was also designed in which the two feature vectors were combined into one and made into a 96-dimensional feature vector. The same learning pattern used in the design was used, the number of category groups was 7, and the number of category reference feature vectors in the category dictionary 42 was 905,590,1255,382,933,1272,1083, a total of 64.
I made 20 pieces. For recognition, the same value was used for each parameter.

As the character pattern for evaluation, 27120 characters of 8 fonts different from the data used for designing each dictionary were used.

As a result of the experiment, a recognition rate of 98.33% was obtained in this example. In addition, the recognition rate of the recognition device, which was used for comparison and recognized by combining two feature vectors into one, was 97.98%. In other words, as in the present embodiment, it is better to identify a plurality of feature vectors individually and perform recognition by integrating the identification results than to perform a recognition process by combining a plurality of feature vectors into one. It can be seen that

As described above, according to the present embodiment, a plurality of types of feature vectors are subjected to identification processing by the separate single feature recognition units 12 to obtain the similarity of each category to each feature vector, and then the category identification is performed. The unit 13 integrates the identification results obtained from each single feature recognition unit using all these similarities,
The final input pattern is recognized. Therefore, it is difficult to use a plurality of types of feature vectors collectively as in the conventional example, "correctly recognizing a pattern that is erroneously recognized by a certain feature vector from the identification result by a different feature vector". It ’s easy to do,
Highly accurate recognition can be realized.

Further, in the conventional method, the time required for recognition increases due to the use of a plurality of types of feature vectors, but in the present embodiment, since each single feature recognition unit performs the identification processing in parallel, the recognition is performed. The time required can be shortened as compared with the conventional example.

FIG. 4 shows a second example of the integrated similarity conversion unit 14 of the present invention.
It is a block diagram which shows the structure of the Example of this concretely. Figure 4
, 51 is a similarity normalization unit, and a single feature recognition unit
The similarity of each category obtained from 12 is normalized by the maximum value of those similarities. Reference numeral 52 denotes an integrated similarity calculation unit, which non-linearly transforms the normalized similarity of each category and obtains the integrated similarity of each category by dividing the non-linearly transformed normalized similarity by its sum value. Is.

Integrated similarity conversion unit configured as described above
The operation of the 14th second embodiment will be described below.

Similar to the first embodiment, first, the similarity normalization unit 51, as shown in (Equation 16), obtains the similarity r _ki (1 ≦ 1) of each category obtained from the corresponding single feature recognition unit 12. k ≦ N
_F : N _F is the number of types of feature vectors, 1 ≦ i ≦ N _C ; N _C is the number of categories), and normalization is performed by dividing by the maximum value r _kmax among these similarities. Similarity s
_Ki (where 0 ≦ s _ki ≦ 1, 1 ≦ i ≦ N _C ; N _C is the number of categories) is output to the integrated similarity calculation unit 52.

[0097]

[Equation 16]

In the integrated similarity calculation unit 52, the similarity normalization unit
The normalized similarity s _ki of each category obtained from 51 is nonlinearly transformed, and the normalized similarity of each nonlinearly transformed category is each normalized (divided) by their sum value, and this is integrated similarity. It is output to the category determination unit 15 as the degree t _ki (where 1 ≦ i ≦ N _C ). In other words, the integrated similarity t _ki is f () when the nonlinear function is

[0099]

[Equation 17]

It can be expressed as The nonlinear function f (x) is 0 ≦ f (x) ≦ 1, f, where 0 ≦ x ≦ 1.
It is a monotonically increasing function that satisfies (0) = 0 and f (1) = 1. For example, there is a function such as (Equation 18). However,
0 ≦ a ≦ 1.

[0101]

[Equation 18]

Although the integrated similarity calculation unit 52 obtains the integrated similarity t _ki for all categories, a plurality of candidates having a large normalized similarity s _{ki in} each category are selected as candidate categories. Individual selection may be performed and integrated similarity may be calculated for these. At this time, the integrated similarity t _ki is ω where the set of candidate categories is ω.

[0103]

[Formula 19]

It can be expressed as In addition, as a method of selection, the normalized similarity may be selected to be a threshold value or more. By the way, by selecting a plurality of candidate categories in this way, the number of categories subjected to non-linear conversion is limited, so that the amount of calculation regarding the integrated similarity can be reduced. Further, normally, only the category having a high degree of similarity affects the recognition performance. Therefore, even if the integrated similarity is calculated only for the candidate categories in this manner, the recognition performance is hardly affected and the recognition processing can be speeded up.

By the way, in each single feature recognition section 12,
When observing the similarity of each category when erroneously recognizing, (1) the magnitude of the maximum similarity is not so related to the erroneous recognition (2) the difference between the maximum similarity and the second largest similarity. It turns out that there are many false recognitions when it is small. Therefore, in the present embodiment, after the similarity obtained from each single feature recognition unit 12 is normalized by the maximum similarity, the process of emphasizing the difference between the maximum similarity and the second largest similarity, that is, the maximum similarity. When the difference between the degree of similarity and the second largest similarity is small, the difference is made smaller and the magnitude of the similarity itself is made smaller, so that a single feature that may cause misrecognition. A process of reducing the reliability (similarity) of the identification result obtained from the recognition unit 12 is performed. Therefore, when the final determination is performed by the category determination unit 15, the result of the other single feature recognition unit 12 becomes important.

Here, the results obtained experimentally are shown. The experiment was the same as in the first example. However, in the integrated similarity calculation unit 52, the nonlinear function used is (Equation 1
8) and a = 0.5.

As a result of the experiment, the recognition rate of 98.38% was obtained in this example, and the recognition rate was improved by 0.05% as compared with the first example.

As described above, according to this embodiment, by performing the above-described conversion on the similarity obtained from each single feature recognizing unit 12, a pattern which is erroneously recognized by a certain feature vector may be a different feature vector. The correct recognition based on the discrimination result by can be performed better than in the first embodiment, and the recognition can be performed with higher accuracy.

FIG. 5 is a block diagram specifically showing the configuration of the second embodiment of the category determination section 15 of the present invention. In FIG. 5, 61 is an input unit for inputting the integrated similarity of each category obtained from the integrated similarity conversion unit 14.
Reference numeral 62 denotes a multi-input / single-output signal processing unit, which is the lower-layer input unit 61 connected to it or the multi-input / single-output signal processing unit 62
Is multiplied by a weighting factor, which is the degree of connection, and the sum is summed to output a threshold value. Here, the plurality of multi-input / single-output signal processing units 62 have a layered structure, are networked so that signals are propagated only to upper layers without mutual coupling in each layer, and thus to a category for an input pattern. A final identification is made. Reference numeral 63 is a maximum value selection unit that selects the maximum value from the outputs of the plurality of multi-input / output signal processing units 62 in the uppermost layer.

FIG. 6 is a block diagram specifically showing the structure of the multi-input / single-output signal processing unit 62. In FIG.
Reference numeral 71 is an input unit for inputting a signal. Reference numeral 72 denotes a weighting coefficient storage unit that stores a weighting coefficient for weighting a plurality of input signals from the input unit 71. A multiplier 73 multiplies the weight coefficient of the weight coefficient storage unit 72 by the input signal from the input unit 71. 74 is an adder, which sums the output values of all the multipliers 73. Reference numeral 75 denotes a threshold processing unit, which limits the output value of the adder 74 to a value within a certain range.

The category determination unit 15 configured as described above
The operation of the second embodiment will be described below.

In the category judgment unit 15, first, the input unit 61
The integrated similarity t _{ki of} each category obtained from each integrated similarity conversion unit 14 is input to. That is, the input unit 61 is prepared by (the number of categories × the number of types of feature vectors),
Integrated similarity t of each category for each feature vector
_{ki (1 ≦ k ≦ N F} ; N F is the number of types of feature vectors, 1 ≦ i
≦ N _C ; N _C is the number of categories) are input to the corresponding input units 61. Each multi-input one-output signal processing unit 62
Is a weighting factor w, which is the degree of connection stored in the weighting factor storage unit 72 and the output of the lower-layer input unit 61 connected to it as shown in FIG. 6, or the multiple-input one-output signal processing unit 62. _ij is multiplied by the multiplier 73, and each multiplier
After the sum of the outputs of 73 is calculated by the adder 74, the threshold value processing unit 75 converts the sum and outputs the output value to the upper layer. That is,
The i-th multi-input one-output signal processing unit 62 of a certain layer shown in FIG.
The output value I _i of the input unit 71 is I _j , the input value connected to the input unit 71, and the weighting coefficient w that is the degree of the connection.
_ij (the connection weight of the i-th multi-input one-output signal processing unit and the j-th input),

[0113]

[Equation 20]

It can be expressed as The input / output characteristics of the threshold processing unit 75 are shown in FIG. For example, the input / output characteristic of the threshold value processing unit 65 that limits the output to the range of (0, 1) is

[0115]

[Equation 21]

It can be expressed mathematically as follows. However, a is an input of the threshold processing unit 65. The threshold processing unit
The input / output characteristic of 75 may be a threshold function other than the above.

The multi-input / single-output signal processing unit 62 in the uppermost layer is
The number of categories of patterns is set, and each multi-input one-output signal processing unit 62 in the uppermost layer corresponds to each category of patterns. The maximum value selection unit 63 selects the maximum output value among the output values of the multi-input one-output signal processing unit 62 in the uppermost layer, and outputs the corresponding category as the final recognition result of the input pattern.

The weighting factor of each multi-input / single-output signal processing unit 62 uses the integrated similarity of each category with respect to each feature vector of the input pattern in advance, and determines the weight of the highest layer corresponding to the category of the input pattern. Multi-input / single-output signal processor
The learning is performed so that 62 produces the maximum output. As a result, the category determination unit 15 can identify the category of the input pattern. The design method of these weighting factors is performed by a learning algorithm called an error back propagation method. For the error backpropagation method, for example, DE Rumelhart, GEHinton and
"Learning Representations by RJ Williams
Back-Propagating Errors, "Nature, vol.323, pp.533
-536, Oct. 9, 1986).

As described above, according to this embodiment, the similarities of all the respective categories obtained from the respective integrated similarity converting units 14 are used, that is, all the similarities obtained as a result of identifying the respective feature vectors are calculated. Then, the final similarity of each category is obtained. Therefore, as compared with the first embodiment, the category can be flexibly identified by using a lot of information, so that the recognition performance with higher accuracy can be obtained.

FIG. 8 is a block diagram specifically showing the configuration of the third embodiment of the category determination section 15 of the present invention. In FIG. 8, reference numeral 91 is an input unit, and a plurality of integrated similarity calculation units 14
The integrated similarities obtained from are combined and input as an integrated similarity vector. Reference numeral 92 denotes an integrated similarity dictionary, which stores a plurality of reference integrated similarity vectors indicating representative values of each category of the integrated similarity vector. A distance calculation unit 93 calculates the distance between the integrated similarity vector and all reference integrated similarity vectors stored in the integrated similarity dictionary 92. A minimum value selection unit 94 selects the minimum value from the distances obtained from the distance calculation unit 93.

The category determination unit 15 configured as described above
The operation of the third embodiment will be described below.

In the category judging section 15, first, the input section 91 collects the integrated similarities obtained from the respective integrated similarity calculating sections 14 into one as an integrated similarity vector T as shown in (Equation 22). It is input and output to N _C (N _C is the number of categories) distance calculation units 93.

[0123]

[Equation 22]

Each distance calculation unit 93 has an integrated similarity dictionary 92.
The reference integrated similarity vector Z _i (1 ≦ i ≦ N _C ) indicating the representative value of the integrated similarity of each category stored in is read out, and the distance d between T and Z _i shown in the equation (23) is read. _i is calculated and output to the minimum value selection unit 94.

[0125]

[Equation 23]

The minimum value selection unit 94 selects the smallest distance among the distances obtained from the distance calculation unit 93, and outputs the corresponding category as the final recognition result of the input pattern.

The reference integrated similarity vector indicating the representative value of the integrated similarity of each category stored in the integrated similarity dictionary 92 is the category of the pattern with respect to the integrated similarity obtained from the input pattern. Distance calculator corresponding to
The output of 93 is designed in advance so as to generate the minimum output as compared with the outputs of the other distance calculation units 93.

The method of designing the reference integrated similarity vector is performed by, for example, a learning algorithm called a learning vector quantization method (LVQ). Regarding the learning vector quantization method, for example, the fine classification unit 23 shown in FIG.
T.Kohone as described in the design method of category dictionary 42 of
"Learning Vector Quantization for PatternRe" by n
cognition ", Helsinki University of Technology, Rep
ort TKK-F-A601 (1986.11).

As described above, according to this embodiment, the similarities of all the respective categories obtained from the respective integrated similarity converting units 14 are used, that is, all the similarities obtained as a result of identifying the respective feature vectors are calculated. Then, the final similarity of each category is obtained. Therefore, as compared with the first embodiment, the category can be flexibly identified by using a lot of information, so that the recognition performance with higher accuracy can be obtained.

In the single recognition unit 12 of this embodiment,
The subclassification unit 23 does not have the configuration shown in FIG. 3, but has an input unit for inputting the feature vector of the input pattern output from the subclassification unit input signal selection unit 25, as shown in FIG. 5, and a layer structure. Alternatively, there may be a plurality of multi-input one-output signal units connected in a network so that signals propagate only to an upper layer without mutual coupling in each layer.

[0131]

According to the present invention, a plurality of types of feature vectors are discriminated by separate single feature recognition units to obtain the similarity of each category to each feature vector, and then the category discrimination unit determines all of these. The final input pattern is recognized by integrating the recognition results obtained from the single feature recognition units by using the similarity of. Therefore, it is difficult to use a plurality of types of feature vectors collectively as in the conventional example, "correctly recognizing a pattern that is erroneously recognized by a certain feature vector from the identification result by a different feature vector". It becomes easy and highly accurate recognition can be realized.

Further, in the conventional method, the time required for recognition increases due to the use of a plurality of types of feature vectors, but in the present embodiment, since each single feature recognition unit performs the identification processing in parallel, the recognition is performed. The time required can be shortened as compared with the conventional example.

[Brief description of drawings]

FIG. 1A is a block diagram showing an embodiment of a pattern recognition apparatus according to the present invention. FIG. 1B is a block diagram showing an embodiment of a single feature recognition unit of the pattern recognition apparatus according to the present invention.

FIG. 2 is a block diagram showing an embodiment of a fuzzy major classification unit of the pattern recognition device according to the present invention.

FIG. 3 is a block diagram showing a first embodiment of a subclassification unit of the pattern recognition device according to the present invention.

FIG. 4 is a block diagram showing a second embodiment of the integrated similarity conversion unit of the pattern recognition device in the present invention.

FIG. 5 is a block diagram showing a second embodiment of a category determination unit of the pattern recognition device according to the present invention.

FIG. 6 is a block diagram showing an embodiment of a multi-input / output signal processing unit of a category judging unit which is a second embodiment of the present invention.

FIG. 7 is an input / output characteristic diagram of a threshold value processing unit in the multi-input / single-output signal processing unit of the category determination unit according to the second embodiment of the present invention.

FIG. 8 is a block diagram showing a third embodiment of the category determination section of the pattern recognition device in the present invention.

FIG. 9 is a block diagram showing an embodiment of a conventional pattern recognition device.

[Explanation of symbols]

 11 Feature extraction unit 12 Single feature recognition unit 13 Category identification unit 14 Integrated similarity conversion unit 15 Category determination unit 16 Similarity normalization unit 17 Adder 18 Maximum value selection unit 21 Group dictionary 22 Fuzzy large classification unit 23 Fine classification unit 24 Group selection unit 25 Fine classification unit Input signal selection unit 26 Single feature similarity calculation unit 27 Multiplier 28 Category similarity calculation unit

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Taiji Yuki, 1006 Kadoma, Kadoma City, Osaka Prefecture Matsushita Electric Industrial Co., Ltd. (72) Koji Yamamoto, 1006 Kadoma, Kadoma City, Osaka Matsushita Electric Industrial Co., Ltd. (72) Inventor Toshio Niwa 1006 Kadoma, Kadoma City, Osaka Prefecture Matsushita Electric Industrial Co., Ltd.

Claims (8)

[Claims] 1. A plurality of feature extraction units that obtain different feature vectors from input patterns, and a plurality of single feature recognition units that obtain a degree of similarity, which is the degree to which the input pattern belongs to each category, from each of the feature vectors. And a category discriminating unit that discriminates the input pattern using the similarity obtained from a plurality of the single feature recognizing units, and the single feature recognizing unit is a set of patterns in which the feature vectors are similar. A group dictionary in which a plurality of group reference feature vectors representing a category group consisting of are stored, and the degree to which an input pattern belongs to each category group using the group reference feature vector and the feature vector of the input pattern. A fuzzy major classification unit that calculates the degree of group membership and each feature pattern of each of the input patterns included in the category group using the feature vector of the input pattern. Based on group selection information obtained from the group selection unit, a plurality of sub-classification units that obtain the degree of similarity within a group that is a degree of belonging to the category, a group selection unit that selects a plurality of category groups from the group membership degree, and Fine classification unit for selecting a fine classification unit for inputting a feature vector of an input pattern, an input signal selection unit, a group membership degree of the category group selected by the group selection unit, and an intra-group similarity obtained by the fine classification unit Using a single feature similarity calculation unit for calculating the similarity of each category with respect to the input pattern, wherein the single feature similarity calculation unit includes the group membership degree of the category group selected by the group selection unit and the fineness. A plurality of multipliers for multiplying all the in-group similarities obtained from the fine classification unit, which are input with the feature vector of the input pattern from the classification unit input signal selection unit, and the output value of the multiplier for each category A large number of A category similarity calculation unit that selects individual pieces and obtains the sum of these output values, wherein the category identification unit calculates the similarity of each category obtained from the single feature recognition unit corresponding to each of the feature vectors. Of a plurality of integrated similarity conversion unit, which is converted to integrated similarity by normalizing with the maximum value of the similarity, and the integrated similarity obtained from the plurality of integrated similarity conversion units A pattern recognition apparatus comprising: a category determination unit that obtains a final similarity and identifies a final input pattern. 2. The integrated similarity conversion unit includes a similarity normalization unit that normalizes the similarity of each category obtained from the single feature recognition unit by the maximum value of the similarity, and the similarity. Non-linear conversion of the normalized similarity of each category obtained from the normalization unit, and the integrated similarity of each category is obtained by dividing the non-linearly converted normalized similarity by its sum value. The pattern recognition apparatus according to claim 1, further comprising a calculation unit. 3. The category determination unit obtains the final similarity of each category by adding the integrated similarity of each category obtained from a plurality of integrated similarity calculation units for each same category. The pattern recognition device according to item 1 or 2. 4. The category determination unit comprises a plurality of multi-input one-output signal processing units having a layered structure, network-connected so that signals are propagated only to upper layers without mutual coupling in each layer, and The multi-input / single-output signal processing unit includes a weight coefficient storage unit that holds a plurality of weight coefficients, an input unit that inputs a plurality of input signals, and a weight coefficient stored in the weight coefficient storage unit from the input unit. A multiplying means for weighting the input signal; an adding means for adding the plurality of input signals weighted by the multiplying means; and a threshold processing section for limiting the output of the adding means to a value within a certain range. The pattern recognition device according to claim 1 or 2, which is characterized in that. 5. The category determination unit has a plurality of reference integrated similarity vectors indicating a representative value of each category of integrated similarity vectors obtained by combining the integrated similarities obtained from a plurality of integrated similarity calculation units. And a plurality of distance calculation units that calculate the distances between the stored integrated similarity dictionary and the integrated similarity vector and all reference integrated similarity vectors stored in the integrated similarity dictionary. Claim 1 characterized by
Or the recognition device described in 2. 6. A fuzzy major classification unit, a plurality of distance calculation units for calculating distances between a feature vector of an input pattern and all group reference feature vectors stored in a group dictionary, and the distance calculation unit. A plurality of dividers for calculating the reciprocal of the output, an adder for adding the outputs of the divider, a plurality of multipliers for multiplying the output of the adder and the output of the distance calculator, The pattern recognition apparatus according to claim 1, further comprising a plurality of dividers that calculate an inverse number of an output of the multiplier. 7. The subclassification unit includes a category dictionary that stores a plurality of category-reference feature vectors that represent representative values of each category of a pattern, and the feature vectors and all categories that are stored in the category dictionary. A plurality of distance calculators for calculating the distance to the reference feature vector, a plurality of dividers for calculating the reciprocal of the output of the distance calculator, and an adder for adding the outputs of the dividers 2. A plurality of multipliers for multiplying the output of the adder and the output of the distance calculation unit, and a plurality of dividers for calculating the reciprocal of the output of the multiplier. The pattern recognition device according to any one of 1 to 6. 8. The subclassification unit comprises a plurality of multi-input / output signal processing units having a layered structure, having no mutual coupling in each layer, and being network-connected so that signals propagate only to an upper layer. The multi-input-output signal processing unit includes a weighting factor storage unit that holds a plurality of weighting factors, an input unit that inputs a plurality of input signals, and a weighting factor stored in the weighting factor storage unit from the input unit. Multiplication means for weighting the input signal,
7. An adding means for adding together a plurality of input signals weighted by the multiplying means, and a threshold processing section for limiting the output of the adding means to a value within a certain range. The pattern recognition device according to any one of claims.