JPH0215103B2 - - Google Patents

Description

[Detailed Description of the Invention] Technical Field of the Invention The present invention relates to an information recognition device that recognizes characters, sounds, etc., and more specifically, the present invention relates to an information recognition device that recognizes characters, sounds, etc. The present invention relates to an information recognition device that can perform information recognition after confirmation.

Background of the technology, prior art, and problems When recognizing information such as text or speech, rather than using a standard dictionary stored in advance in the recognition device, it is necessary to add or change specific recognition words. It has been proposed to improve the recognition rate by, for example, registering a unique recognition target word for a specific user (see Japanese Patent Application No. 41995/1982 and Japanese Patent Application No. 45262/1983).

There are also devices that allow the user to change the recognition dictionary for target words that are misrecognized or unrecognizable while using the recognition device; If you simply change the target word, the recognition rate will improve for that target word, but the change will have a negative effect on the relationship with other target words in the recognition dictionary, and the overall recognition rate may decrease. be. Furthermore, unilaterally adding target words that have caused misrecognition is of course limited by the limited memory capacity of the recognition dictionary; Increasing the number of items causes a responsiveness problem in that the recognition time increases.

In order to improve recognition performance, it is necessary to add erroneously recognized or unrecognized target words as feature patterns (dictionary patterns) to the recognition dictionary, and to add unnecessary or unrecognized target words to the recognition dictionary. It must be possible to remove patterns and optimize the recognition dictionary.

One way to determine how much influence a dictionary pattern can have on recognition results is to count within the device the frequency with which the dictionary pattern is used for correct recognition, and to delete dictionary patterns that have caused erroneous recognition. There are some methods that have made this possible (for example, see Japanese Patent Application No. 111007-1982). However, this type of method has a problem in that it takes a relatively long period of time to produce results after a certain period of use. In addition, since the user determines whether the recognition result is correct or not, the recognition device has a means for detecting how the user reacts to the recognition result presented to the user, and displays the recognition result until the user responds. It is necessary to have a means for storing which dictionary pattern the recognition result is based on. Therefore, there is a problem that the device becomes complicated.

However, adding or deleting a dictionary pattern not only affects the recognition rate for the target recognition word as described above, but also affects other categories that have a certain relationship with the recognition word. There is a need for an apparatus that can effectively change the recognition dictionary in consideration of such circumstances.

Purpose of the Invention The purpose of the present invention is to solve the above-mentioned problems of the prior art and make it possible to change, ie add, integrate, or delete unnecessary target words to the dictionary patterns of specific target words in order to improve the recognition rate. It is an object of the present invention to provide an information recognition device capable of improving the recognition rate of the device as a whole by improving the recognition rate of other target words by making changes such as addition or deletion while maintaining the recognition rate of other target words.

Another object of the present invention is to provide an information recognition device that can maintain appropriate responsiveness without complicating the device configuration.

The above object of the present invention is to extract all target words in the recognition dictionary that may be affected by the change of the target word when changing the recognition dictionary for a specific target word, and to This is achieved based on the concept of optimizing the trade-off between the device and the user by letting the user decide whether or not a good result can be achieved in terms of recognition. In the following, the term "change" means not only the change itself, but also additions and deletions.

Structure of the Invention In the present invention, a feature pattern is extracted from input image information, the distance between the feature pattern and a dictionary pattern stored in a dictionary is determined, and the above-mentioned input is made into the category represented by the dictionary pattern with the smallest distance. When changing one of the dictionary patterns, the pattern recognition device recognizes that the dictionary pattern belongs to the dictionary pattern, and when changing one of the dictionary patterns, the distance calculation means calculates and obtains a dictionary pattern whose distance from the dictionary pattern is within a certain value; related category image information input requesting means for displaying categories of dictionary patterns within a certain value and requesting the user to input image information for the categories of the displayed dictionary patterns; a pre-change recognition means that recognizes a recognition result for the input image information based on the dictionary pattern before the change; and a post-change recognition means that recognizes a recognition result for the image information input in response to the request using the dictionary pattern after the change. , a comparison display means for comparing and displaying the results recognized by the dictionary patterns before and after the change; a change necessity indicating means for confirming whether or not to make the change and inputting an instruction; A pattern recognition device is provided, comprising a dictionary change control means for controlling change of a dictionary according to an instruction from an instruction means.

Embodiment of the Invention An embodiment of the present invention will be described below with reference to the accompanying drawings.

FIG. 1 illustrates a recognition device 1 based on the present invention that performs handwritten character recognition as an example of information recognition. In FIG. 1, the recognition device 1 includes a function specifying section 10 for inputting various information for character recognition or dictionary modification, a tablet 11 for inputting handwritten characters, a control section 12, and a display for displaying recognition results. It has a display 13, such as a cathode ray tube display, for displaying information for interaction with other users, and a recognition section 14. The control unit 12 includes a control circuit 121 that performs various controls and processes described below in response to signals from the function specifying unit 10, and an input buffer 122 that inputs handwritten characters from the tablet 11 based on control signals from the control circuit 121. , a dictionary pattern save buffer 123 and a display 13, which will be described later.
It is equipped with a display buffer 124 for outputting data to.

The recognition unit 14 receives an input signal from the tablet 11 from an input buffer 122 and generates a feature pattern using a feature creation circuit 141, an output signal S41 from the feature creation circuit, and a signal S from the control circuit 121.
21 and selectively outputs one of the signals based on the control signal of the control circuit;
It has a matching circuit 143 that compares and matches the selected output with the recognition target word stored in the recognition dictionary storage section 144 and other processes that will be described later, and a recognition dictionary storage section 144.

The operation of the apparatus shown in FIG. 1 will be described below. Initial values of recognition dictionaries such as categories "A", "B", "C", etc. are stored as dictionary patterns in the recognition dictionary storage unit 144 as A ₀ , B ₀ , C _{0 ,} etc. as illustrated in FIG. shall be taken as a thing. The feature space of the dictionary pattern is
Although it is not actually possible to express it two-dimensionally as shown in the figure, FIGS. 2 to 4 are schematically shown two-dimensionally for ease of explanation. Figure 2 shows only _the categories A, B, _and C out of all the categories to _be recognized _. , ₀ , ₀ , indicating that they are adjacent.

Input "A ₁ " is input from tablet 11,
If the input “A ₁ ” is found to be in the area ₀ of the category “A” of the recognition dictionary by the comparison of the matching circuit 143, even if it is far from the standard pattern A ₀ , the distance from A ₀ is determined. It is recognized as "A" because it is the closest to other B ₀ , C _{0 ,} etc. However, if the position of the character entered by the user with the intention of "A" in the feature space is actually "A 2 ", then the position in the feature space of the character input by the user with the intention of "A" is actually "A ₂ ", then the distance from the center of category A, that is, A ₀ , is rather than the distance from category B. The input "A ₂ " is near the center B ₀ of the area ₀ , and the recognition result is "B".

In such a case, if the above input "A ₂ " is the user's common font and he wants it to be recognized as "A", then the category "A" should be set to include "A ₂ " as well. It is necessary to change the recognition dictionary. This modification will be described below with reference to the flowchart of FIG. In this example, a case of addition as a change will be explained, but the same process will be applied to other changes.

The user instructs addition of a dictionary pattern via the function specifying section 10 (step 10). The recognition unit 14 calculates the distance between the dictionary pattern for the selected category "A" and other dictionary patterns (step 11). Next, the categories of dictionary patterns whose distances are within a certain value, as shown in FIG. 3, are displayed on the display 13 because the categories "B" and "C" are within a certain distance ( Step 12). Furthermore, a message indicating that the user should input handwriting is displayed on the display 13 (step 13). Examples of these displays are shown in FIG. 6A.

Dictionary pattern "A ₂ " that the user wishes to add
Using Tablet 11, write a few letters for the categories “A”, “B”, and “C” related to
14). Based on the recognition dictionary, the device 1 calculates the recognition status when the dictionary pattern _A2 to be added is not added to the category "A", when it is added, that is, before the change, and when it is changed. Display 1 as shown in Figure 6B.
3 displays the results (steps 15 to 17).

Here, the change in recognition before and after the change will be described with reference to FIG. Assuming that "A ₂ " is simply added to the category "A", the intermediate position between the new "A ₂ " and the center B ₀ of the category "B" will be the boundary between the categories "A" and "B". . Therefore, the area of the new category "A" will be expanded to ₀₊ ', while the area of category "B" will be decreased to ₁ .

Therefore, whether it is better to keep the dictionary in the state before the change as shown in Figure 2 or to change it as shown in Figure 3 depends on what kind of input characters will be input from now on. It must be determined whether it is more advantageous to increase the recognition rate by keeping For this purpose, it would be more effective to compare the recognition rates of a large number of input character forms in the dictionary state, but since such a thing cannot be done in real time, it is difficult for the user to do anything on the spot. By having the user fill in the character shape of a character and displaying a comparison of the recognition results as shown in Figure 6B, the user can use the information to decide whether to change the dictionary or leave it as it was. It provides:
This allows the user to instruct whether or not the change is necessary (step 18).

The user indicates the judgment result (step 18).

The dictionary pattern save buffer 123 in FIG. 1 stores the information retrieved from the dictionary storage section 144, input patterns to be added, etc. at the time of addition, deletion, or modification as described above, and stores the information retrieved from the dictionary storage section 144 and input patterns to be added. In response to the instruction, the recognition dictionary is determined, and the dictionary storage section 144 is updated based on it (step 19).

Figure 4 shows the category diagram of the recognition dictionary after it has been updated sequentially.By deleting the dictionary pattern of _A0 for category "A", region ₁ is deleted from region ′ ₊₁ . The area of category “A” becomes ′ and category “C”
This is an example of the case where the area expands to ₀ + ₁ . The change operation for this can also be performed in the same way as the addition.

As described above, the user can create or update the recognition dictionary so that his own character expressions that have a low recognition rate or are likely to be unrecognizable are recognized in accordance with his own characters. In addition, through optimization such as deleting common characters that are far from one's own character pattern and focusing on strengthening one's own error-prone parts, the recognition rate can be improved and responsiveness can be improved, in other words, recognition time can be shortened. can be achieved.

Although the above example was given for one English character, it can be done similarly for any other character or word.

The present invention can take various modifications other than those described above. For example, in the above embodiment, character information was described as the information to be recognized, but the same process can be applied to other information such as graphic information, voice input information, etc. When performing voice input information, use Tablet 1 in Figure 1.
Change 1 to microphone and input buffer 12
2. The feature generation circuit 141 and the like may be replaced with circuits that can handle voice input; in fact, those already used as voice recognition devices may be used.

Effects of the Invention According to the present invention, it is possible to create a recognition dictionary that matches the individuality of the user using the most appropriate feature space assignment based on the user's understanding through dialogue with the user. Therefore, it is possible to improve the recognition rate of not only the updated recognition word but also the recognition rate of the entire device. Additionally, since it becomes easier for the user to learn how to write letters in order to be recognized correctly, it also has the effect of training the writing style to naturally match the device. In addition, since the dictionary is not automatically changed by the device but at the discretion of the user, the user can confirm whether or not the change can be made, which can lead to mistakes or meaningless dictionary patterns due to incorrect changes. Things will go away.

Furthermore, in updating the recognition dictionary in the present invention, by extracting all objects affected by the update and delegating the decision to the user, the desired recognition dictionary can be updated without error without complicating the device configuration. can do.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the configuration of a recognition device as an embodiment of the present invention, FIGS. 5 is a flow chart showing one aspect of the operation of the device shown in FIG. 1, and FIGS. 6A and 6B are diagrams showing one form of interactive information displayed on the display in the device shown in FIG. 1 based on the present invention. (Explanation of symbols) 10...Function designation unit, 11...Tablet, 12...Control unit, 121...Control circuit, 1
22...Input buffer, 123...Evacuation buffer, 1
24...display buffer, 13...display, 14
... Recognition section, 141 ... Feature creation circuit, 142 ... Input switching circuit, 143 ... Verification circuit, 144 ... Recognition dictionary storage section.

Claims (1)

[Claims] 1. Extract a feature pattern from input image information, find the distance between the feature pattern and a dictionary pattern stored in a dictionary, and place the input into the category represented by the dictionary pattern with the smallest distance. When changing one of the dictionary patterns, the pattern recognition device recognizes that the dictionary pattern belongs to the dictionary, and the distance calculation means S11 calculates and obtains a dictionary pattern whose distance from the dictionary pattern is within a certain value; Related category image information input request means S1 that displays categories of dictionary patterns within a certain value and requests the user to input image information of the categories of the displayed dictionary patterns.
2, S13; a pre-change recognition means S15 that recognizes the recognition result for the image information input in response to the above request based on the dictionary pattern before the change; A changed recognition means S16 that recognizes based on a dictionary pattern, and a contrast display means S that compares and displays the results recognized using the dictionary patterns before and after the change.
17, a change necessity instructing means S18 for confirming whether or not to perform the change and inputting an instruction; and a dictionary change control means S19 for controlling dictionary changes in accordance with the instruction from the change necessity instructing means S18.
A pattern recognition device comprising: