JPH0222784A - Device for preparing dictionary for character recognition - Google Patents

Abstract

PURPOSE:To prepare a high-quality dictionary for character recognition which is hardly affected by adverse influences when the dictionary is updated with an inferior character picture by causing a dictionary updating section to always add the feature quantity of a text at a fixed rate. CONSTITUTION:A character feature input section 3 inputs the feature quantity of a certain picture of a character C as F=(f1, f2-fn). A dictionary updating section 4 calculates a new standard feature quantity G''=(g1'', g2''-gi''-gn'') from an equation gi''={gix(S-1)+fi}/S by using the standard feature quantity of the character C, G=(g1, g2-gn), stored in a standard feature quantity storing section 1 and the feature quantity F=(f1, f2-fn) inputted from the input section 3 and stores the calculated G'' in the storing section 1 as the standard feature quantity of the character C. The S is the number of so far updated time and corresponds to the number of character picture used for updating the standard feature quantity.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to a recognition target used in a character recognition device that recognizes printed and handwritten characters from newspapers, magazines, etc. and converts them into information such as JIS codes. ρ also relates to a character recognition dictionary creation device that creates and changes a dictionary that stores information for identifying characters.

BACKGROUND ART In a conventional character recognition dictionary creation device configured as shown in FIG. 12, the dictionary stores only the standard features of each character, and the dictionary is updated as shown in FIG. 13. , the standard feature quantity gi stored in the dictionary 11 and the new character feature ■
The weighted sum of Sofi and the number of individual characters registered in the dictionary S
(For example, Japanese Unexamined Patent Publication No. 59-85579).

Problems to be Solved by the Invention However, the above method has the following problems. For example, for each pixel composing a character image, there are M or more (including the pixel of interest) in each direction shown in FIG. 3(b).
(M is set in advance)) Check whether or not they are continuous, assign a direction code to the pixel of interest, check the connectivity of pixels for each direction code, extract strokes, and use the number of strokes in each direction as the feature quantity of the character. This will be explained using a case as an example. Figure 4 (a),
When the characteristic amounts of partially blurred character images such as the character images shown in (b), (c), and (d) are determined, they are as shown in FIG. 6. When the standard features shown in Figure 5 are updated using the conventional processing method shown in Figure 13 using such a defective character image, the standard features become as shown in Figure 14, and even with only a slight update, This results in moving away from the correct rFJ feature amount. Therefore, for example, it becomes difficult to distinguish between the standard feature amounts of rEJ and "F", and erroneous recognition is likely to occur.

In view of these points, the present invention has been developed to provide a character image that will not have an adverse effect such as, for example, causing the dictionary to incorrectly recognize the correct character when the dictionary is erroneously updated due to a defective character image as described above. The purpose of this invention is to provide a recognition dictionary creation device.

Means for Solving the Problems In order to solve the above problems, the character recognition dictionary creation device of the present invention includes a dictionary that stores standard feature amounts that are standard feature amounts of each character, and a dictionary that stores standard feature amounts that are standard feature amounts of each character. A model feature storage unit that stores model feature quantities; a character feature input unit that inputs feature quantities of arbitrary characters; and a character feature input unit that inputs feature quantities of arbitrary characters; Add the feature amount of an arbitrary character input in the character feature In input section and the feature amount of the model character to be updated stored in the model feature storage section at a specified ratio, and add the feature amount to the dictionary. and a dictionary update unit that updates the stored standard feature quantities.

In addition, there is also a dictionary that stores standard features that are the standard features of each character and sample features of each character, a character feature input section that inputs the features of any character, and Dictionary update that updates the standard feature amount stored in the dictionary by adding the feature amount of any character input in the character feature input section to the stored standard feature amount of the character to be updated at a specified rate. It has a section.

Operation According to the first configuration of the present invention, the feature values of the model are always added at a constant rate in the dictionary update section, so that when the dictionary is updated, the feature values of defective character images are added to the character feature input section. Even when the standard feature is inputted in , it is less likely that the standard feature will be erroneously updated to a large extent.

Further, according to the second configuration described above, in the dictionary updating section, the standard feature amount in the dictionary is changed, but the feature amount of the example in the dictionary is not changed. Even if the feature amounts are inputted in the character feature input section, the dictionary will not misrecognize the correct characters.

Embodiment FIG. 1 is a block diagram of a character recognition dictionary creation device in a first embodiment of the present invention.

In FIG. 1, 1 is a dictionary that stores standard features that are standard features of each character, 2 is a model feature storage unit that stores model features of each character, and 3 is an arbitrary number. 4 is a character feature finger input section for inputting the feature amount of the character, and 4 is the standard feature amount of the character to be updated stored in the dictionary 1, and the feature amount of any character inputted in the character feature input section 3 and the above. This is a dictionary updating section that updates the standard feature amount stored in the dictionary 1 by adding the feature amount of the model of the character to be updated stored in the model feature storage section 2 at a specified ratio.

The operation of the character recognition dictionary creation device of this embodiment configured as described above will be described below.

It is assumed that each character has n feature amounts.

In the character feature input section 3, the feature amount of the image with the character C is F= (fl, f2. @ *, fl,
* Enter as @, fn). The dictionary update unit 4 calculates the standard feature amount G of the character C stored in the dictionary 1 = (gl+ g2+ . . . , gl, @
・+gn) and the feature amount F=l, r2. inputted in the character feature amount input section 3. * II, fl
, @ 11. fn) and the model feature amount T of the character C stored in the model feature storage unit 2 = (tl, t2. * *,
tl, @6. tn) and equation (1)
The new standard feature quantity G = = (gl
−+ g2−+ ・・+gl−+ ・・+ gn″
) is calculated, and the standard feature ff1G- is stored in the dictionary 1 as the standard feature m of the character C. S is the current number of updates, and corresponds to the number of character images used to update the standard feature amount.

gl-:tlXa+(glX(S-1) 10 fl)/
S X (1-α) (1) (1≦1nn) α is a preset value, and 0≦α≦1.

FIG. 2 shows the flow of dictionary update processing. Here, for each pixel constituting the character image, it is checked whether or not M or more (M is preset) including the pixel of interest are connected in each direction shown in FIG. 3(b), and a direction code is assigned to the pixel of interest. An example will be explained in which the strokes are extracted by checking the connectivity of pixels for each direction code, and the number of strokes in each direction is used as the feature amount of the character. For example, the direction of the stroke extracted for the character image shown in Figure 3(a) is shown in Figure 3(a).
C).

FIG. 5 shows the standard feature values of each character stored in the dictionary 1, and FIG. 7 shows the model feature values stored in the model feature storage unit 2.

The feature amounts of partially blurred character images such as those shown in FIGS. 4(a), (b), (c), and (d) are as shown in FIG. 6. Using the feature quantities of these character images in order, the fifth
When the standard feature amount shown in the figure is updated according to the flow shown in FIG. 2, it becomes as shown in FIG. 8 (assuming α=0.4).

As described above, according to this embodiment, by providing a dictionary update unit that adds the model feature amount at a certain rate when updating the standard feature amount of the dictionary, the standard feature amount is updated by a defective character image. In case 7, the changes will not be so large that they deviate from the correct character feature values.In addition, the calculation method in the dictionary update section is such that the model feature values are added at a certain rate. Needless to say, the method using equation (1) used in this embodiment is not the only method.

Further, FIG. 9 is a block diagram of a character recognition dictionary creation device in a second embodiment of the present invention. In FIG. 9, reference numeral 5 denotes a dictionary, which includes a standard feature storage unit 6 that stores standard feature quantities that are standard feature quantities of each character, and a model feature storage unit 7 that stores model feature quantities. Become. Reference numeral 3 denotes a character feature input section for inputting the feature amount of an arbitrary character, which is similar to that shown in FIG. 8 is a standard feature obtained by adding the feature of an arbitrary character input in the character feature input section 3 to the standard feature of the character to be updated stored in the standard feature storage 7 of the dictionary 5 at a specified ratio. This is a dictionary updating unit that updates standard feature quantities stored in the feature quantity storage unit 7.

The operation of the character recognition dictionary creation device of this embodiment configured as described above will be described below.

It is assumed that each character has n feature amounts.

In the character feature input section 3, the feature amount of the image containing the character C is set as F = (fle f21 11 *, fl,
I*, fn). The dictionary update unit 8 updates the standard feature ■standard feature amount G” (gL g2+ ” ”+ gl+ of the character C stored in the storage unit 6)
” ” + gn) and the feature ■F= (fl, f2. @ @,
fl, @e, fn) using equation (2) to create a new standard feature amount G”=(gl”+g2”)
, . . +gl''', . S is the current number of updates, and corresponds to the number of character images used to update the standard feature amount. What is changed by the dictionary updating unit 8 is only the contents of the standard feature storage unit 6 in the dictionary w5.
The contents of the model feature amount storage section 7 within are not changed.

gl”= (g+x(S −1)−+−r+) /S
...(2) (1≦i:an) Here, to explain the feature amount used in the first example as an example, the contents of the dictionary shown in FIG. 10 are the four characters shown in FIG. When updated using the feature amount, the result becomes as shown in FIG. 11.

As described above, according to this embodiment, even if the standard feature values in the dictionary are updated due to a defective character image and deviate from the correct character feature values, the model feature values that remain unchanged are stored. Therefore, the inconvenience that the correct character is recognized incorrectly does not occur.

It goes without saying that the calculation method used in updating the standard feature amount is not limited to the method using equation (2) used in this embodiment.

DETAILED DESCRIPTION OF THE INVENTION As described in detail, according to the present invention, it is possible to create a high-quality character recognition dictionary with less adverse effects when the dictionary is updated with defective character images.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a dictionary creation device for character recognition according to a first embodiment of the present invention, FIG. 2 is a flowchart of calculations in the dictionary updating section of the same embodiment, and FIG. 3 is a character feature in the same embodiment. FIG. 4 is an explanatory diagram of the character image given to the character feature input section in the same embodiment, and FIG. 5 is an explanation showing the standard feature quantities stored in the dictionary before updating in the same embodiment. 6 is an explanatory diagram showing the feature amount of an arbitrary character obtained in the character feature input section in the same embodiment, FIG. 7 is an explanatory diagram showing a sample feature amount in the same embodiment, and FIG. The figure is an explanatory diagram showing the standard features stored in the updated dictionary in the same embodiment, FIG. 9 is a block diagram of the character recognition dictionary creation device in the second embodiment of the present invention, and FIG. An explanatory diagram showing the contents of the dictionary before updating in the same embodiment,
FIG. 11 is an explanatory diagram showing the contents of the dictionary after updating in the same embodiment, FIG. 12 is a block diagram of a character recognition dictionary creation device according to an embodiment of the conventional technology, and FIG. 13 is an example of the conventional embodiment. FIG. 14 is an explanatory diagram showing the dictionaries stored in the updated dictionary in the same embodiment. 1...Dictionary, 2...Example feature amount storage unit, 3.Fist/character feature input unit, 4...Dictionary update unit, 5.φ fist dictionary,
6. Fist/standard feature storage section, 7. Model feature storage section, 3e Kensha dictionary update section. Name of agent Patent attorney Shigetaka Kurino Haka 1 person Page 5 Sentence*a) ?L・-Sei1!Feature amount j=−°Ishi t’s sentence 9j Feature amount 几・Segin number number

Claims (2)

[Claims] (1) A dictionary that stores standard features that are the standard features of each character, a model feature storage that stores model features of each character, and a dictionary that stores the standard features of each character. A character feature input unit to be input, a standard feature of a character to be updated stored in the dictionary, a feature of any character input in the character feature input unit, and a standard feature of the character to be updated stored in the model feature storage unit. A dictionary creation device for character recognition, comprising: a dictionary update unit that updates standard feature quantities stored in the dictionary by adding feature quantities of model characters of characters to be updated, respectively, at specified ratios. (2) a dictionary that stores standard feature quantities that are standard feature quantities of each character and model feature quantities of each character; a character feature input unit that inputs feature quantities of arbitrary characters; and the dictionary The standard feature amount stored in the dictionary is updated by adding the feature amount of any character input in the character feature input section to the standard feature amount of the character to be updated stored in the dictionary. A dictionary creation device for character recognition, comprising a dictionary updating section.