JPH0359471B2 - - Google Patents

Description

TECHNICAL FIELD OF THE INVENTION The present invention relates to a character recognition device, and more particularly to a character recognition device equipped with a personal dictionary.

Prior Art and its Problems Conventional character recognition devices recognize or process input characters by comparing them with a personal dictionary. As a method for creating this personal dictionary, there is a method in which the dictionary contains both a general-purpose dictionary template and a template constructed from personal character features. However, such a method has the drawback that the number of templates increases, and a high recognition rate cannot be obtained in proportion to the increase in the number of templates.

OBJECTS OF THE INVENTION In view of the drawbacks of the prior art, it is an object of the present invention to provide a character recognition device equipped with a personal dictionary that achieves a high recognition rate with a small dictionary capacity.

Structure of the Invention According to the present invention, the object is to provide a personal dictionary that uses a general-purpose dictionary as an initial dictionary, a matching circuit that matches input personal characters with the personal dictionary, and character codes and input character characteristics of full-screen recognition results. a full-screen recognition result temporary storage section for storing the full-screen recognition result temporary storage section; a correction circuit for correcting erroneous recognition results in the full-screen recognition result temporary storage section; and a correcting circuit for correcting incorrect recognition results in the full-screen recognition result temporary storage section; According to the weight information of the template of the personal dictionary, the following criteria are met: (a) Among the templates of the personal dictionary that are in the same category as the input character and have the minimum weight, find the one that is closest to the input character feature; The input character information and its weight are grouped as (P, p), and the template belonging to the character category and its weight are (T ₁ , t ₁ ), (T ₂ ,
t ₂ ), ... (T _L , t _L ), and when the one closest to the input character information (P, p) among the templates is shown as ( _TV , t _V ), the template (T _V , t _V ) is replaced by (t _V T _V +p ^P /t _V +p, t _V +P). This is accomplished by providing a device.

Embodiments of the Invention Examples of the present invention will be described in detail below with reference to the drawings.

The present invention focuses on the fact that the properties of general-purpose dictionary templates and personal character features are essentially the same, and utilizes this to achieve a high recognition rate with a small number of dictionary templates. A dictionary template is constructed by weighting and averaging each character feature.

Specifically, we used the algorithm described below.

A general-purpose dictionary is set as an initial dictionary when personal characters are not input. Then, weights are assigned to each of the general-purpose dictionary template and individual character features. For example, the former may be set as 1 and the latter as 2. Then, when a personal character is input, the dictionary template is updated based on the character characteristics. This method will be explained with reference to the flowchart of FIG. 1 with character category C fixed. In other words, input personal characters and dictionary templates belong to category C. A universal dictionary template is provided before personal characters are entered into the device. Let L be the upper limit of the number of templates in a dictionary belonging to character category C, and let H be the number of initial states (number of templates in a general-purpose dictionary).

In block 1, input personal character features (P,
If L>H in block 2, the characteristics of the L-H characters belonging to character category C that are first input into the device are sequentially added to the dictionary as templates (block 3).

Now, assume that the number of templates belonging to character category C is L and personal characters of the same category are input. Here, the characteristics of the character including the template and its weight m are combined and expressed as (M, m).

Then, in block 4, the templates belonging to character category C are (T ₁ , t ₁ ), (T ₂ , t ₂ ), ...,
(T _L , t _L ), find the minimum value of the weights {t ₁ , t ₂ , ..., t _L }, and assume that the value is t _i1 = t _i2 = ... = t _ik . Next, in block 5, the one closest to the input character feature (P, p) among T _i1 , T _i2 , ..., T _ik is selected as T _V
Then, in block 6, the template (T _V , t _V ) in Block 7 is replaced with (t _V T _V +p ^P /t _V +p, t _V +p).

The dictionary is updated in this way based on the criterion that ``for any template, the same number of templates and personal character features in the general dictionary that are close to each other in the feature space are averaged''. The reason for setting such a criterion is to reflect the distribution of characteristics of character category C with a small number of templates. Note that if a character feature is composed of several small features, the personal dictionary is updated for each small feature by considering the weight of each small feature.

Next, one embodiment of the present invention will be explained with reference to the block diagram of FIG.

In FIG. 2, an input character 11 is replaced with binary information in an observation section 12, and then the size of the binary pattern is normalized in a normalization circuit 13. The normalized pattern is processed by the feature extraction circuit 14.
The input character features thus obtained are sent to the matching circuit 16. Personal dictionary 1
5 is updated using a general-purpose dictionary as an initial dictionary, as will be described later. This personal dictionary 15 is compared with input character features in a matching circuit 16. The recognition result is stored in one screen recognition result temporary storage unit 1.
7 and displayed on the display 18. The character of the input character 11 is recognized, the recognition result is displayed on the entire screen of the display 18, and the character code and the characteristics of the input character are stored in a one-screen recognition result temporary storage section.

In response, the operator corrects the recognition result using the keyboard 19 while looking at the display 18.
That is, the character code of the recognition result of the read character in the one-screen recognition result temporary storage section 17 is corrected by a command from the keyboard 19, and the result is displayed on the display 1.
Display on 8. When the correction for one screen is completed in this way, the operator issues a command from the keyboard 19 to the apparatus to move to the next page. Then, the personal character feature to which the correct character code is attached in the one-screen recognition result temporary storage section 17, the template of the same character category in the personal dictionary, and its weight are sent to the personal dictionary modification circuit 20 (,).

Next, a detailed block diagram of the one-screen recognition result temporary storage section 17 is shown in FIG. 3, and a data flow for explaining its operation is shown in FIG. 4.

In FIG. 3, 21 is a template number determination circuit, 22 is a template weight ascending circuit, and 23 is a template number determining circuit.
2 shows a matching circuit, and 24 shows a weighted averaging circuit. Next, the flow of the data is shown in FIG. First, information input 31 from the storage section 17
If there is, the judgment in block 2 of FIG. 1 is made in the template number judgment circuit, and if H>L, the input personal character information is sent to the personal dictionary 15. If H=L, the information is sent to the template weight ascending circuit 22 to perform the process of block 4 in FIG. 1, and to the matching circuit 23 to perform the process of block 5 in FIG. The output of the matching circuit 23 is sent to the weighted averaging circuit 2.
4, where it undergoes the processing of block 6 in FIG. 1 and is sent to the personal dictionary 15. At the output of the averaging circuit 24, the content 33 of the template is replaced with the averaged content 34. In this way, the personal dictionary is automatically modified according to the characteristics of one screen's worth of personal characters.
When the change is completed, the next character is automatically read from the observation unit 12.

Effects of the Invention As explained in detail above, according to the present invention, a personal dictionary consisting of a small number of effective templates can be created from a general-purpose dictionary template and personal character features, and a high recognition rate can be achieved with a small capacity. effective.

[Brief explanation of drawings]

FIG. 1 is a flowchart showing the algorithm of the device of the present invention, FIG. 2 is a block diagram showing one embodiment of the present invention, and FIG. 3 is a diagram of the personal dictionary modification circuit in the block diagram of FIG. A detailed block diagram, FIG. 4, is a diagram showing the data flow in the block diagram of FIG. In the figure, 11 is the input character, 12 is the observation part, 1
3 is a normalization circuit, 14 is a feature extraction circuit, 15 is a personal dictionary, 16 is a collation circuit, 17 is a one-screen recognition result temporary storage section, 18 is a display, 19 is a keyboard, 20 is a personal dictionary modification circuit, 21 2 shows a template number determination circuit, 22 a template weight ascending circuit, 23 a matching circuit, and 24 a weighted averaging circuit.

Claims (1)

[Scope of Claims] 1. A personal dictionary whose initial dictionary is a general-purpose dictionary, a collation circuit that collates input personal characters with the personal dictionary, and a full-screen recognition result character code and input character characteristics. A screen recognition result temporary storage unit, a correction circuit for correcting erroneous recognition results in the full screen recognition result temporary storage unit, weight information of input characters using correctly coded input character features, and a personal dictionary template. According to the weight information of the following criteria: (a) Find the one closest to the input character feature among the personal dictionary templates in the same category as the input character and having the minimum weight, and (b) find the input character information and The weights are expressed as a set (P, p), and the templates belonging to the character category and their weights are (T ₁ , t ₁ ), (T ₂ ,
t ₂ ), ... (T _L , t _L ), and when the one closest to the input character information (P, p) among the templates is shown as ( _TV , t _V ), the template (T _V , t _V ) is replaced by (t _V T _V +p ^P /t _V +p, t _V +P). Device.