JP2000306045A - Word recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a word recognizing device for recognizing a handwritten character string improved in recognition even in the tilt or different character size of a word. SOLUTION: This word recognizing device is provided with an individual character recognizing part 21 for recognizing respective characters constituting the word of an input image, a word recognizing part 22 for collectively recognizing the word of the input image, a certainty calculating part 10 for calculating the certainty of the word from the recognized result of the individual character recognizing part 21 and calculating the certainty of the word from the recognized result of the word recognizing part 22 and a certainty comparing part 11 for comparing the certainty calculated results of the individual character recognizing part 21 and the word recognizing part 22 calculated by the certainty calculating part 10 and on the basis of the compared result of the certainty, the certainty comparing part 11 selects greater one of the certainty of the individual character recognizing part 21 and the word recognizing part 22.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a word recognition device for recognizing a handwritten character string.

[0002]

2. Description of the Related Art In recent years, a character recognition device OCR (Optical Character) has been used as a handwritten character input device.
Reader) demand is increasing. Among the character recognition devices, in addition to the individual character recognition device, a word recognition device that focuses on a character string to improve character recognition has appeared.

FIG. 7A shows a conventional individual character recognition device 31.
FIG. The individual character recognition device 31 includes an image input unit 2, a normalization unit 3, a feature extraction unit 4, a feature comparison unit 5, a recognition result output unit 6, and an individual character dictionary 7.

[0004] The image input unit 2 acquires an input image of handwritten characters using a scanner or the like. The normalizing unit 3 makes the size and position of the characters uniform. The feature extraction unit 4 extracts a feature amount from the normalized input image. The feature matching unit 5
The feature amount of the character in the individual character dictionary 7 is compared with the feature amount extracted by the feature extraction unit 4. The recognition result output unit 6 outputs the result of comparison by the feature comparison unit 5 to a display or the like.
The individual character dictionary 7 is composed of characters and their feature amounts.

The operation will be described with reference to an address example shown in FIG.

The character string “Tokyo” input from the image input unit 2 is adjusted to a prescribed size by the normalization unit 3, and the feature amount of the character is extracted by the feature extraction unit 4. The feature amount of “East”, the feature amount of “K”, and the feature amount of “Tokyo” of the extraction result are passed to the feature matching unit 5.

The feature matching unit 5 compares all the characters in the individual character dictionary 7 with the feature amounts of “East”, “Kyo”, and “To” from the feature extraction unit 4, and as a result, the difference between the feature amounts is calculated. The character having the minimum value is output from the recognition result output unit 6 as a recognition result.

FIG. 8 (a) shows a configuration diagram of a conventional word recognition apparatus 1.

The word recognition apparatus 1 includes an image input unit 2, a normalization unit 3, a feature extraction unit 4, a feature comparison unit 5, a feature synthesis unit 8, a recognition result output unit 6, a word list unit 9, and an individual character dictionary 7. .

The image input unit 2 acquires an input image of a handwritten character using a scanner or the like. The normalizing unit 3 makes the size and position of the characters uniform. The feature extraction unit 4 extracts a feature amount from the input image. The feature synthesizing unit 8 selects a word, extracts feature amounts of individual characters constituting the selected word from the individual character dictionary 7, and synthesizes them. The feature matching unit 5 compares the feature amount of the word combined by the feature combining unit 8 with the feature amount of the input image extracted by the feature extracting unit 4. Word list part 9
Consists of multiple word lists. The recognition result output unit 6
A word having the most similar feature amount among the results matched by the feature matching unit 5 is output to a display or the like. The individual character dictionary 7 includes characters and their feature amounts.

FIG. 8B shows an example of an address.

It is assumed that characters serving as keys such as prefectures are printed at predetermined coordinates.

The image input unit 2 receives the input image
First, a position where a character serving as a key of a prefecture is printed is detected, and a character string “Tokyo” preceding the position is cut out. The extracted character string is normalized by the normalization unit 3. The feature amount of the normalized word is extracted by the feature extracting unit 4 and passed to the feature matching unit 5. The feature synthesizing unit 8 selects a word list of prefectures in the word list unit 9. In the word list, “Hokukai”, “Tokyo”, “Okinawa”, and the like are registered as words, and the individual character dictionary 7 registers feature amounts of all characters.

First, the feature synthesizing section 8 first selects the word "North Sea" according to the word list of prefectures, and extracts the feature amounts of "North" and "Sea" from the individual character dictionary 7,
The feature quantity of "North Sea" is synthesized. The result of the synthesis is compared with the feature amount of the input image by the feature matching unit 5, and a difference in feature amount between the two is obtained. Then, the same processing is performed for all the words in the word list. When the matching process is completed, a word having a small difference in the feature amount among the words is output to the recognition result output unit 6.
And output as a recognition result.

[0015]

With an apparatus that individually recognizes characters, it has been difficult to accurately extract a character string in contact with the character. For this reason, a word recognition method has been developed which collectively captures contact characters as words without separating them. However, in the method of recognizing words collectively without separating them into individual characters, since the word feature amount is extracted from the image of the word, high recognition can be performed if the characters are arranged at the same height. If the is skewed or if the letters that make up the word vary in size,
Recognition accuracy was low.

Also, when synthesizing the feature amounts of words in the feature synthesizing section of the word recognition device, the synthesizing is performed based on the individual character dictionary. Since the synthesis process is performed for the "Uzumasa" portion including the column every time, the synthesis process is useless.

Further, when the feature amount of the word is synthesized by the feature synthesizing unit of the word recognition device, the difference between the feature amount of the input image and the feature dimension of the input image is reduced due to the extraction from the single-size individual character dictionary. This caused misreading. Skewed words also caused misreading.

In the feature collating unit of the word recognition device, misreading is likely to occur between words containing the same character due to the feature amount of the same character. An object of the present invention is to provide (1)
(3) The same as the provision of a word recognition device having an improved recognition accuracy for a tilted word and a word having a different character size, and (2) the provision of a word recognition device having a feature synthesizing method in which the synthesis speed is high and the recognition accuracy is improved. It is an object of the present invention to provide a word recognition device having a feature matching method that prevents misreading between words including characters.

[0019]

An individual character recognizing unit for recognizing individual characters constituting words of an input image, a word recognizing unit for collectively recognizing words of the input image, and a recognition result of the individual character recognizing unit Calculation unit for calculating the certainty factor of the word from the words and calculating the certainty factor of the word from the recognition result of the word recognition unit, and the confidence of the individual character recognition unit and the word recognition unit calculated by the certainty calculation unit A word recognition device having a certainty comparison unit that compares the calculation results, the certainty comparison unit selecting one of the individual character recognition unit and the word recognition unit with a higher certainty based on the comparison result of the certainty. .

With this configuration, by using the individual character recognition unit and the word recognition unit together, it is possible to achieve higher recognition accuracy than when each is used alone. That is, the word recognizing units complement each other such that the recognition accuracy of the contact character is high, and the individual recognizing units are high in the recognition accuracy of the inclination character and the change in the character size.

Further, an individual character recognizing unit for recognizing individual characters of the input image, a word recognizing unit for recognizing words of the input image collectively, and detecting an interval between individual characters constituting words of the input image. A character spacing detecting means for comparing the spacing detected by the character spacing detecting means with a predetermined threshold, wherein the character spacing comparing means recognizes the word if the detected character spacing is smaller than the predetermined threshold. Select the department,
A word recognition device configured to select an individual character recognition unit if the detected character spacing is greater than a predetermined threshold.

With this configuration, by using the individual character recognizing unit and the word recognizing unit separately, it is possible to achieve higher recognition accuracy than when each is used alone, and it is possible to perform processing at high speed. In other words, if the character spacing is narrow, use a word recognition unit to increase the recognition accuracy of contact characters.If the character spacing is wide, use an individual recognition unit to improve the recognition accuracy of tilted characters and different character sizes. Complement each other, such as making them higher.

Further, the word recognition apparatus performs word recognition by comparing the result of extracting the feature of a word from an input image with the result of synthesizing the feature of a word from the feature of each character in the dictionary. And a dictionary that holds a plurality of characters and partial character string feature amounts and a plurality of various character size images, and synthesizes a word feature amount from the character and partial character string feature amounts based on the dictionary. And a feature synthesizing unit having means for synthesizing a word image from a character image based on the dictionary and extracting a feature amount.

According to this configuration, not only the characters constituting the word but also the feature amount of the partial character string having a high frequency of appearance are previously stored in the dictionary, so that the speed of synthesizing the feature amount of the word can be increased, and different character sizes can be obtained. , Having a partial character string enhances recognition accuracy. Further, by performing the image synthesis, the feature amount of the inclined word can be extracted, so that the recognition accuracy of the inclined word is improved.

Also, a word recognition apparatus having a feature synthesizing unit having a selecting means for selecting a character for synthesizing a characteristic amount of a predetermined word size and a characteristic amount of a partial character string from a dictionary in a size corresponding to the number of characters.

With this configuration, by combining a plurality of different character sizes in accordance with the number of characters and combining the feature amounts of words with the same normalized size, the number of feature dimensions of the input image and the number of combined feature dimensions are the same. And the recognition accuracy is improved.

Also, a word having a feature synthesizing unit having a conversion means for converting a character constituting a word extracted from the dictionary and a feature of a partial character string into a character of a predetermined size and a feature of a partial character string. Recognition device.

According to this configuration, since the feature amounts of words of different sizes can be synthesized based on the feature amounts of the standard character size and partial character string size for individual character recognition, recognition accuracy can be improved with a small dictionary capacity.

Further, a complexity weighting means for weighting the characteristic amount of a character or a partial character string according to the complexity of a character or a partial character string constituting a word extracted from the dictionary according to a dose, and a complexity weighting means. A word recognition device having a feature synthesizing unit having a character synthesizing unit for converting a character constituting a word extracted from the dictionary and a partial character string into a character of a predetermined size and a partial character string based on the weighted data.

According to this configuration, by increasing the feature dimension of a character having a complicated shape and decreasing the feature dimension of a character having a simple shape, recognition according to the shape of the character can be achieved. Can be recognized.

An inclination detecting means for detecting the inclination of a word in the input image, and an image composing the word are extracted from a dictionary, and the word images are arranged along the inclination detected by the inclination detecting means to synthesize the images. And a feature synthesizing unit having means for extracting a feature amount.

With this configuration, it is possible to generate a feature amount of a word corresponding to the inclination of the input word, and the recognition accuracy of the word in the inclined character string is improved.

Also, an input image weighting means for weighting the features of characters and partial character strings constituting words of the input image, and a feature synthesis for weighting the features of characters and partial character strings generated by the feature synthesis unit. A word recognition device having a configuration further comprising a feature matching unit having matching means for matching the feature amount extracted by the weighting means and the input image weighting means with the feature amount extracted by the feature synthesis weighting means.

According to this configuration, for a word having a specific character or a specific partial character string constituting a word, the weight of the characteristic amount of the same part is reduced, and the characteristic amount of another character or a partial character string is reduced. , The difference between the feature amounts of similar words becomes clear, and the recognition accuracy is increased.

[0035]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1A shows a configuration diagram of a word recognition apparatus 1 according to a first embodiment.

The image input unit 2, individual character recognition unit 21, word recognition unit 22, certainty calculation unit 10, certainty comparison unit 11, recognition result output unit 6, and individual character dictionary 7.

The individual character recognition unit 21 includes a normalization unit 3, a feature extraction unit 4, and a feature collation unit 5.

The word recognition unit 22 includes a normalization unit 3, a feature extraction unit 4, a feature comparison unit 5, a feature synthesis unit 8, and a word list unit 9.
It is composed of

The individual character recognition section 21 and the word recognition section 22 are the same as in the conventional example. The certainty calculating unit 10 calculates the certainty of each character constituting the word from the individual character recognizing unit 21, calculates the average, and calculates the certainty of the word from the word recognizing unit 22.
The certainty comparison unit 11 compares the certainty of the individual character recognition unit 21 with the certainty of the word recognition unit 22. The certainty means that the similarity with the feature amount of the input image is high.

This is an apparatus that uses both individual character recognition processing and collective word recognition processing. Handwritten characters input by the image input unit 2 are recognized by each recognition unit for characters and words. Then, the degree of certainty of the recognition result is obtained.

The method of calculating the certainty factor is as follows.

First, a feature vector indicating a feature amount of an input image is extracted. Next, the individual character recognition unit 21 obtains a feature vector indicating a feature amount of a character in the dictionary, and obtains a distance value of a difference from the feature vector of the input image. Do this for all the letters that make up the word.

Further, the word recognition device 1 obtains a feature vector indicating the feature amount of the word synthesized by the feature synthesizing unit 8, and obtains a distance value of a difference from the feature vector of the input image.

Next, using a large number of samples, the distribution of the distance values is examined in advance, and the distribution is normalized so that the range of the distribution is within the range of 0 to 1. Convert distance values to distance values on the distribution. The distribution is f
If expressed as (X), the certainty factor R (d) when the distance value is d is expressed by the following equation. The distance value d is 0 ≦ d
≦ 1, and the case where d = 1 is regarded as the maximum similarity.

[0045]

[Formula 1]

For example, in the case of a two-character word, the individual character recognition unit 21 outputs R, which is the average of the certainty factors for each character.
(D1) + R (d2) / 2 is obtained. Next, the word recognition unit 22
Then, R (d3) is obtained from the certainty factor of the word. The certainty comparison unit compares the magnitudes. As a result, a recognition result having a high degree of certainty is output from the recognition result output unit 6 as a recognition result.

FIG. 2 is a block diagram of the word recognition device 1 of the second embodiment.
Shown in

It comprises an image input unit 2, a character interval detection unit 12, a threshold value comparison unit 13, an individual character recognition unit 21, a word recognition unit 22, a recognition result output unit 6, and an individual character dictionary 7.

The individual character recognition unit 21 includes a normalization unit 3, a feature extraction unit 4, and a feature collation unit 5.

The word recognition unit 22 includes a normalization unit 3, a feature extraction unit 4, a feature comparison unit 5, a feature synthesis unit 8, and a word list unit 9.
It is composed of The individual character recognition unit 21 and the word recognition unit 22 are the same as in the conventional example. The character interval detection unit 12 detects an interval between characters of an input image. The threshold comparing unit 13 compares the interval detected by the character interval detecting unit 12 with a threshold.

The processing according to the second embodiment is used when a high processing speed is required because a long processing time is required to pass two recognition processings. First, the character spacing of the input word is detected, and two recognition processes are selectively used depending on whether the character spacing is larger or smaller than a threshold.

The operation will be described with reference to FIG.

Assuming that the image input from the image input unit 2 is Wakayama, the character spacing detecting unit 12 detects the character spacing. For example, a circumscribed rectangle of a character is obtained from the connected black pixels, and a space between the circumscribed rectangles is detected as a character interval. If the detection result is smaller than the threshold value, there is a high possibility that the characters are close to each other, so that the word recognition unit 22 performs the processing. If the detection result is
If it is larger than the threshold value, it is determined that they are not close to each other, and the individual character recognizing unit 21 performs the processing for each individual character.

FIG. 3 is a block diagram of the word recognition device 1 according to the third embodiment.
Shown in

The word recognition device 1 comprises an image input unit 2, a normalization unit 3, a feature extraction unit 4, and a recognition processing unit 17. The recognition processing unit 17 includes a feature collating unit 5, a feature synthesizing unit 8, a recognition result output unit 6, a word list unit 9, an individual character dictionary 7, and a word dictionary 16.

The image input unit 2 obtains an input image of a handwritten character from a scanner or the like and detects the inclination of a word. The normalizing unit 3 makes the size and position of the characters uniform. The feature extraction unit 4 extracts a feature amount from the input image.
The feature synthesizing unit 8 selects a word, extracts a feature amount of each character constituting the selected word from the individual character dictionary 7,
The feature amounts of the partial character strings constituting the word are extracted from the word dictionary 16 and synthesized. In addition, character and partial character string size conversion, complexity conversion, inclination processing, and the like are performed. The feature matching unit 5 compares the feature amount of the word combined by the feature combining unit 8 with the feature amount of the input image extracted by the feature extracting unit 4. The word list section 9 includes a plurality of word lists. The recognition result output unit 6 outputs, to a display or the like, a word having the most similar feature amount among the results matched by the feature matching unit 5. The individual character dictionary 7
It is composed of characters, their features, and their images. Word dictionary
16 is composed of a common partial character string of a compound word and its characteristic amount.

The operation of the word recognition device 1 will be described with reference to FIGS.

First, the case where "Tokyo" is input as an input image will be described with reference to an example of an address shown in FIG.

It is assumed that key characters such as prefectures are printed at specified coordinates. The image input unit 2 first detects a position where a character serving as a key of a prefecture is printed from the input image, and cuts out a character string in front of the position. Next, the connected black pixel is extracted to obtain a rectangle circumscribing it. Set a straight line to the center of all the obtained circumscribed rectangles. Then, the inclination is set as the inclination of the word, and is compared with a predetermined threshold.

The description will be made separately for the case where the inclination is smaller than the threshold and for the case where the inclination is greater than or equal to the threshold. (1) In the case of inclination <threshold The normalization unit 3 converts the input character string “Tokyo” into
For example, when the normalization to 64 × 64 pixels is completed, the feature amount of the input image “Tokyo” is extracted by the feature extraction unit 4. In the word list of prefectures, words of prefectures such as Hokkai, Tokyo, and Okinawa are registered, and in the individual character dictionary 7, feature amounts of all characters are registered.

Next, the processing flow of the recognition processing section 17 is shown in FIG.
(A).

Although the description is for individual characters, the same applies to partial character strings.

First, the feature synthesizing unit 8 first searches for the word "Hokukai" according to the word list. (Step S11)
Next, the "north" feature amount and the "sea" feature amount are extracted from the individual character dictionary 7 and the feature amounts are synthesized, and there are three types of methods. Feature synthesis processing is performed such that the dimension number of the feature vector after the feature synthesis is equal to the dimension number of the normalization of the input image.

A case where a mesh feature is used as a feature will be described. The mesh feature amount is a feature amount in the small rectangle obtained by dividing the number of black pixels in each small rectangle by the small rectangle size when the image is divided into mxn meshes. For example, if the normalized size is 64 × 64 pixels, the size of the small rectangle divided into 8 × 8 mesh is 8 × 8. Assuming that there are 32 black pixels in a certain small rectangle, the mesh feature amount is 32/64 = 0.5. The input image is 64 × 64
Since the feature amount is extracted by normalizing the pixels with 8 × 8 mesh, the feature synthesis result is also synthesized so that the 64 × 64 pixel becomes 8 × 8 mesh.

The first method is feature synthesis using an individual character dictionary 7 having a reduced character size. (Step S121 in FIG. 4). An explanatory diagram is shown in FIG.

This is used for an OCR-dedicated machine or the like that can provide a large memory capacity for a dictionary at high speed.

As for one character and a partial character string, feature amounts extracted from a plurality of character sizes are provided as individual character dictionaries 7 respectively. An example using an individual character dictionary 7 having reduced characters of 1/2 size is shown. A half size reduced character has an 8 × 4 mesh.

"Hokumi" based on the word list of prefectures
Are extracted from the individual character dictionary 7 and are combined to obtain the feature quantity of the word “Hokukai” having an 8 × 8 mesh size. (Step S13) The word feature amount is compared with the feature amount of the input image, and the difference is stored. (Step S14 in FIG. 4) The second method is feature synthesis of the character size conversion method.
(Step S122 in FIG. 4) This is used for a portable information terminal or the like which cannot have a large memory capacity for a dictionary.

This method is a method of generating feature quantities of various character sizes by applying conversion coefficients to feature quantities of a standard character size, and synthesizing word feature quantities. FIG. 5B shows an explanatory diagram of the feature amount synthesis.

A feature amount of a reduced character size is synthesized by applying a conversion coefficient to a feature amount of a standard character size.

For example, since the feature quantity of the standard character size of 64 × 64 pixels is 8 × 8 mesh, the standard character size of “north” is multiplied by 8 × 4 conversion coefficient to １ ／. The feature value of the character of the size "north" is obtained.
In addition, by multiplying the standard character size of “sea” by a conversion coefficient of 8 × 4, a feature amount of a character of “sea” of サ イ ズ size can be obtained. The combination of the two is used as the word feature of “Hokukai”. (Step S13 in FIG. 4) This word feature is compared with the feature of the input image and the difference is stored. (Step S14 in FIG. 4) The third method is feature synthesis of the complexity conversion method. (FIG. 4
Step S123) The memory capacity for the dictionary, such as a portable information terminal, cannot be large, and is used for a device that requires high recognition accuracy.

In order to perform processing with high accuracy, first, the complexity of characters constituting a word is checked. As the complexity, for example, an explanatory diagram of the linear density calculation using the vertical line density is shown in FIG.
It is shown in (c). This is a method of combining the word feature amounts by weighting the feature amounts according to the ratio of the sum of the linear densities of the characters.

For example, for the word "north sea", "north"
If the line density of the sea is 30 and the line density of the “sea” is 50, the character size at this time is 64 × 64 pixels, so that it is divided into 8 × 8 meshes. The number is 8 × 30 / (30 + 50) = 3, and the feature of “sea” is
The number of horizontal meshes is set to 8 × 50 / (30 + 50) = 5. And
By multiplying the standard character size of “north” by a conversion coefficient of 8 × 3, the feature amount of the character of “north” can be obtained. or"
By multiplying the standard character size of “U” by a conversion coefficient of 8 × 5, the feature amount of the character of “U” can be obtained. The combination of the two is used as the word feature of “Hokukai”. (S in FIG. 4
13 Step) This word feature is compared with the feature of the input image, and the difference is stored. (Step S14 in FIG. 4) These processes are performed for all the words, and finally the feature amount having the minimum difference between the feature amount of the input image and the feature amount from the feature synthesizing unit 8 is regarded as the recognition result. Output. (Step S15 in FIG. 4) Although not shown, the words include Uzumasazuka,
There are many compound words that contain the same substring, as in Uzumasaki. Therefore, instead of synthesizing “Uzumasa” from “Fat” and “Hata” every time when synthesizing feature quantities, for “Uzumasa”, for example, a feature quantity of 文字 character size is extracted from the word dictionary 16 in advance. Then, the characteristic amount of the character of “horse” of 1/4 size and “mound” of 1/4 size is extracted from the individual character dictionary 7, and the word characteristic amount is calculated from the characteristic amount of “Uzumasa” + “horse” + “mound”. Is a method of synthesizing. (2) In the case where the inclination degree ≧ the threshold value.

FIG. 6 shows a recognition processing unit 1 in the case where the inclination degree ≧ the threshold value.
7 is an explanatory diagram.

FIG. 6A shows a processing flow, and FIG. 6B shows an example of an address.

After extracting the character string of the input image "Tokyo" in the address example, the normalization unit 3 normalizes the input image as a sloping word, and then the feature extraction unit 4 performs feature extraction. Next, the word "north sea" is acquired from the word list of prefectures, and the characters constituting the word are used as keys, and the "north"
And get the image of "sea". (Step S51) The acquired images are arranged along the inclination θ. (Step S52) Next, the image is normalized to the same size as the normalized size of the input image. (Step S53) Feature extraction is performed based on the normalized word image. (Step S54) The extracted feature amount is compared with the feature amount extracted from the input image, and the difference is stored. (Step S55) The above processing is performed for all words in the word list. Of all the words, the one with the minimum difference is output from the recognition output unit. (Step S56) Although not shown, in the feature matching unit 5, for example, if the input word is an address and it is known in advance that it is any of "Uzumasa OO", "Uzumasa" Part "is not a valid feature for matching. Therefore, the difference between the feature amounts of similar words is clarified by making the weight of the feature amount of this part low or masking it, and making the weight of the other components high or performing collation while keeping it as it is. Thereby, recognition accuracy is improved.

For example, if the input image is “Uzumasa Mazuka”, this character string is cut out, and the common character portion “Uzumasa” is masked, and the feature is extracted only from the remaining characters “Matsuka”. Then, "Uzumasa" is extracted from the word list. For example, if the extracted word is "Uzumasagakiuchi",
Mask "Uzumasa". Then, the feature amounts of the remaining “Kakiuchi” portion are combined. Then, the feature amount of the input image is compared with the feature amount obtained by the feature synthesis. This collation is performed for all the compound words including Uzumasa in the word list, and the one having the smallest difference in the feature amount is specified as the input word.

[0078]

According to the present invention, word recognition is possible even if the word is inclined, and recognition is possible even if the characters constituting the word are of different character sizes.

Further, by synthesizing feature amounts by selecting a character size according to the number of characters, word recognition accuracy can be improved.
Furthermore, even for a tilted word, the word recognition accuracy is improved by extracting a feature amount by image synthesis. Further, by preliminarily holding the common partial character string or character feature in the case of a compound word, the processing at the time of feature synthesis can be speeded up.

Further, by weighing the feature amount between words consisting of the same character or the same partial character string, misreading can be reduced.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a word recognition device according to a first embodiment.

FIG. 2 is a configuration diagram of a word recognition device according to a second embodiment.

FIG. 3 is a configuration diagram of a word recognition device according to a third embodiment.

FIG. 4 is an explanatory diagram of a recognition processing unit according to a third embodiment when the inclination degree is smaller than a threshold value.

FIG. 5 is an explanatory diagram of feature amount synthesis according to a third embodiment.

FIG. 6 is an explanatory diagram of a recognition processing unit according to a third embodiment when the inclination degree ≧ the threshold value;

FIG. 7 is a configuration diagram of a conventional individual character recognition device.

FIG. 8 is a configuration diagram of a conventional word recognition apparatus.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Word recognition device 2 Image input part 3 Normalization part 4 Feature extraction part 5 Feature collation part 6 Recognition result output part 7 Individual character dictionary 8 Feature synthesis part 9 Word list part 10 Confidence calculation part 11 Confidence comparison part 21 Individual character Recognition unit 22 Word recognition unit

Claims (8)

[Claims] An individual character recognition unit for recognizing individual characters constituting words of an input image, a word recognition unit for collectively recognizing words of an input image, A confidence calculation unit that calculates the confidence and calculates the confidence of the word from the recognition result of the word recognition unit; and calculates the individual character recognition unit and the confidence calculation result of the word recognition unit calculated by the confidence calculation unit. A word recognition device, comprising: a certainty comparison unit for comparing, wherein the certainty comparison unit selects the larger one of the individual character recognition unit and the word recognition unit based on the result of the comparison of the certainty factors. 2. An individual character recognizing unit for recognizing individual characters of an input image, a word recognizing unit for recognizing words of the input image collectively, and detecting an interval between individual characters constituting words of the input image. A character spacing detecting means; and a character spacing comparing means for comparing the spacing detected by the character spacing detecting means with a predetermined threshold, wherein the character spacing comparing means recognizes the word if the detected character spacing is smaller than the predetermined threshold. A word recognition apparatus, wherein a unit is selected, and an individual character recognition unit is selected if the detected character interval is larger than a predetermined threshold. 3. A word recognition apparatus for recognizing a word by comparing a result of extracting a feature of a word from an input image with a result of synthesizing a feature of a word from a feature of an individual character in a dictionary. A dictionary that holds the feature values of a plurality of characters and partial character strings of various sizes and images of the plurality of character sizes, and synthesizes the feature values of words from the feature values of the characters and partial character strings based on the dictionary. A word synthesizing unit having means for synthesizing a word image from a character image based on the dictionary and extracting a feature amount. 4. A feature synthesizing unit having a selecting means for selecting from a dictionary according to claim 3, a character for synthesizing a feature of a predetermined word size and a feature of a partial character string in a size corresponding to the number of characters. The word recognition device according to claim 3, wherein: 5. A feature synthesizing unit having a conversion unit for converting a character constituting a word extracted from the dictionary according to claim 3 and a characteristic amount of a partial character string into a character of a predetermined size and a characteristic amount of a partial character string. The word recognition device according to claim 3, further comprising: 6. A complexity weighting means for weighting a characteristic amount of a character or a partial character string according to the complexity of a character or a partial character string constituting a word extracted from the dictionary according to claim 3. And a conversion unit configured to convert characters and partial character strings constituting words extracted from the dictionary according to claim 3 into characters and partial character strings of a predetermined size based on the weights calculated by the degree weighting unit. The word recognition device according to claim 3, further comprising a unit. 7. An inclination detecting means for detecting an inclination of a word in an input image, and an image forming a word is extracted from the dictionary according to claim 3, and a word image is extracted along the inclination detected by the inclination detecting means. 4. The word recognition device according to claim 3, further comprising a feature synthesizing unit having means for arranging and synthesizing images and extracting feature amounts. 8. An input image weighting means for weighting a feature amount of a character or a partial character string constituting a word of an input image, and a feature amount of a character or a partial character string generated by the feature synthesizing unit according to claim 3. The image processing apparatus according to claim 1, further comprising a feature matching unit having matching means for matching the feature amount extracted by the feature combining weighting means and the feature amount extracted by the feature combining weighting means. 3. The word recognition device according to 3.