JPH0363777B2 - - Google Patents

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to a character selection method, and in particular to a character selection method that performs optimal identification processing for each of a plurality of specific scribes registered in advance, in an identification unit of a character reading device. It concerns the selection method.

[Prior art]

Pattern recognition performed in character reading devices and the like is an operation that associates a character pattern with a category to which it belongs, and the basic process of pattern recognition is shown in FIG.

In FIG. 1, the first-stage observation unit 2 is a functional unit that reads characters on a paper using optical or magnetic means in a character reading device, and captures a given pattern into the recognition system. In the second stage, the preprocessing unit 3 selects a partial pattern of interest (segmentation 5) and shapes the partial pattern to make it easier to extract essential information from the read character pattern. , operations to separate and remove unnecessary information (normalization 6 and noise removal 7) are performed. In the next third stage, the feature extraction section 4, in order to concretely express the characteristics specific to the pattern, these are represented by a small number of numerical values called feature parameters. One of the feature parameters is expressed by a feature vector X. In the next fourth stage, the determination unit 10 and the rank processing unit 50 use the values of the extracted feature parameters (feature vectors If the category has a rank, rank processing is performed and the determined category is output.

The character selection method is related to the stage after the determination unit 10 in FIG. Conventionally, this type of character selection method in character reading devices involves selecting standard character shapes that serve as sample characters in order to remedy rejects caused by multiple candidate categories, or character shapes that are generally easily described with high frequency. It gave priority to subcategories and gave high rankings to those subcategories.

FIG. 2 shows the relationship between subcategories and ranks as described above, taking character category "7" as an example. Here, the priority is set higher for the letters on the left, which are more elaborate or are generally written more frequently. (Hereinafter, character categories will be referred to as categories.) FIG. 3 shows an example of a conventional method of this type. 10 is a feature extraction unit 4 in the character reading device
This is a determination unit that compares and verifies the identification feature vector X created in the above with the identification logic for each subcategory.
Reference numeral 20 denotes a dictionary memory, which stores identification logic 21 corresponding to subcategories and categories/categories to which the subcategories belong.
It consists of a code 22. 30 is a rank memory, which stores rank information associated with the arrangement of subcategories in the dictionary memory 20, and has identification logic L ^A _a for the m-th subcategory.
The run R ^A _a corresponds to (X), and the rank R ^B _b corresponds to the identification logic L ^B _b regarding the nth subcategory (X). A candidate category register 40 stores the candidate category 41 obtained from the determination processing section 10 and its rank 42. 50
is a rank processing unit that selects and outputs a candidate with a high rank in the candidate category register 40.

The operation of the above embodiment will be described below. First, after initializing the candidate category register 40 to Nu, the determination processing unit 10 sets the feature vector
and each identification logic L(X) on the dictionary memory 20
are compared and verified sequentially. Here, the discrimination logic is a function of the feature vector, where the input pattern is X, and the discrimination logic of the subcategory belonging to category A is L ^A
When set to (X), the following judgment is made.

When L ^A (X)=1, X∈A …(1) When L ^A (X)=0, X∈/A …(2) ∈ is a symbol of a set, and X is a 1 of set A. When there are two elements (elements), it is expressed as XεA; otherwise, it is expressed as Xε/A. When L ^A _a (X) = 1 in the process of a series of comparisons and collations, the category code A to which the subcategory it means belongs and the rank in the rank memory 30 specified by the address pointer 23: R ^A _a is stored in the candidate category register 40 via the determination unit 10. Further, when L ^B _b (X)=1 for the same character, category code B and rank: R ^B ' _b are added and stored in the candidate category register 40 in the same way.
At this time, if the identification logic of a plurality of subcategories in the same category satisfies the condition (becomes 1), only the highest rank is stored in the candidate category register 40. Next, in rank processing 50,
Based on the rank content 42, the category with the highest rank is selected from the candidate categories 41 as follows and output.

When R ^A _a < R ^B _b , X∈A…(3) When R ^A _a > R ^B _b , X∈B…(4) When R ^A _a = R ^B _b , Reject…(5) Follow Therefore, even if the discrimination logic conditions are satisfied for multiple categories at the same time, by referring to the rank information, priority is given to subcategories of glyphs that are close to the sample character or that are generally easy to write, and a single It is possible to select candidate categories. In this type of system, ranks are set for an unspecified number of scribes, so it is difficult to read characters with high precision in counter operations, turnaround systems, etc. that handle written data of many scribes. It is advantageous to do so. However, in jobs where specified scribes frequently use character reading devices, the writing tendencies of individuals are almost constant and the related subcategories are limited. If the reading accuracy is different, there is a drawback that the reading accuracy is lowered. There is also a method of providing a dictionary memory consisting of individually designed identification logic for each specific scribe using the data actually written, but this method has the disadvantage that the dictionary memory becomes large in size.

[Purpose of the invention]

The purpose of the present invention is to improve such conventional drawbacks without increasing the dictionary memory size.
Moreover, it is an object of the present invention to provide a highly accurate and economical character selection method that can utilize personal writing tendencies for specific scribes.

[Summary of the invention]

The above-mentioned object of the present invention is to provide a character recognition device that associates a character pattern with a category to which the character pattern belongs using a common dictionary. A memory and a standard rank memory having a rank value set for an unspecified number of unregistered scribes are provided, and prior to character reading, a personal code is input by key operation or information written on an input form. Then, based on the personal code, select the character by switching between referring to the personal rank memory for registered specific scribes and the standard rank memory for unregistered scribes. This is achieved by a character selection method characterized by:

[Embodiments of the invention]

FIG. 4 is a block diagram of a character selection section showing an embodiment of the present invention. 10 is a judgment unit; 20 is a dictionary memory containing identification logic 21 for each subcategory and category codes 22 to which these subcategories belong; 30 is a standard that stores rank values set for an unspecified number of scribes. rank memory, 3
1 and 32 are the first and M-th personal rank memories among M pieces prepared for registering the rank value of each specific scribe, and 40 is the category code 41 that has become a candidate in the determination section 10. a candidate category register that temporarily stores the rank 42; 50 a rank processing unit 60 that selects and outputs a candidate category with a high rank in the candidate category register 40;
70 is a key operation input section for inputting a personal code 61 corresponding to the scribe prior to character reading, and 70 is a key operation input section for selecting one of a plurality of rank memories based on the personal code 61, and determining the contents of the selected rank memory at 10. This is a rank memory selection unit that outputs to. In this embodiment, the contents of the standard rank memory 30 are registered at the same time as the contents of the dictionary memory 20, and the M personal rank memories are arbitrarily set for each specific scribe. It can be registered and changed. Here, the standard rank memory is referred to as 0-rank memory, the personal rank memory is referred to as 1-M-rank memory, and the rank memory selection unit 70 is referred to as 0,
One rank memory is allocated and selected according to any one of the personal codes 1, . . . , M.

The operation of this embodiment based on the above configuration will be described below. Key operation input section 60 prior to character reading
The personal code 61 input from is, for example, 0,
Only (M+1) characters of 1, . . . , m, . . . M are accepted, and the others are ignored. Next, the operation of the rank memory selection section 70 is such that when no personal code is input or when 0 is input, the operation is 0 - rank memory 3.
Only rank values starting from 0 can be accepted, and when a personal code m is obtained, only rank values from the m-rank memory can be accepted only if it is a registered personal code. for example,
Personal code “1” registered as personal code 61
If the 1-rank memory is obtained, the 1-rank memory is selected, and it becomes possible to read characters using the rank memory dedicated to that particular scribe. Similar to the conventional embodiment shown in FIG. 2, the determination unit 10 compares and verifies the identification feature vector X with the identification logic 21 for each subcategory. In this process, if L ^A _a (X)=1, the determination unit 10 obtains the category code: A to which the subcategory meant by the identification logic belongs from the dictionary memory 20 and stores it in the candidate category register 40 . At this time, the address pointer 23 points to the address corresponding to the subcategory on the 0 to M-rank memory,
The rank memory selection unit 70 obtains the rank R ^A ′ _a of the corresponding address from the acceptable 1-rank memory and stores it in the candidate category register 40 . Furthermore, if L ^B _b (X)=1 for the same character, the corresponding category code: B and rank: L ^B ' _b are similarly added and stored in the candidate category register 40. next,
The rank processing unit 50 selects a candidate category with a higher rank and outputs it as a reading result, as in the conventional embodiment shown in FIG.

The contents of the personal rank memory in this embodiment are as follows:
For example, based on a small number of character data written by a specific scribe to be registered, a learning means prepared separately from the character reading device is used to detect subcategories in the dictionary memory that have a high probability of involving those characters. The rank information is determined in such a way that priority is given to them, and the content settings are determined in advance by a registration method in which the rank information is stored in a character reading device via a storage medium such as an FDD (floppy disk). It is something to be carried out.

With this configuration, when reading characters that take into account the ranks of subcategory priorities, characters that are rejected because multiple candidate categories are obtained are read by scribes registered in the personal rank memory. Characters can be selected by switching between referring to a dedicated personal rank memory for scribes and a standard rank memory for an unspecified number of unregistered scribes.

〔Effect of the invention〕

As explained above, according to the present invention, rank information for preferentially handling related subcategories is individually stored according to the writing tendency of a specific scribe,
Because it is structured so that it can be referenced selectively, even if a particular scribe is different from the average and general writing tendency of an unspecified number of scribes, highly accurate character reading by referring to dedicated rank information can be done in the dictionary memory. It has the advantage that it can be realized with a simple configuration without changing the . Also, for a specific scribe whose writing tendency is almost the same as the general writing tendency, rank information is provided in advance in which the rank of a subcategory related to a character shape that is rarely written by the scribe is relatively lowered. This has the advantage that even more accurate character reading can be achieved.

[Brief explanation of drawings]

Figure 1 is a block diagram showing the basic process of pattern recognition, Figure 2 is a diagram of the relationship between character shapes and ranks of subcategories related to category "7", Figure 3 is a block diagram showing the conventional character selection method, and Figure 4 is a block diagram showing the basic process of pattern recognition. The figure is a block diagram of a character selection method showing an embodiment of the present invention. 10... Judgment unit, 20... Dictionary memory, 21...
...Identification logic for each subcategory, 22...Category code, 23...Address pointer, 30...
...Standard rank memory, 31, 32...Personal rank memory, 40...Candidate category register, 41
... Candidate category code, 42 ... Rank information,
50...Rank processing unit, 60...Key operation input unit, 61...Personal code, 70...Rank memory selection unit.

Claims (1)

[Claims] 1. In a character recognition device that uses a common dictionary to match a character pattern with a category to which the character pattern belongs, a personal rank memory having a rank value set exclusively for registered specific scribes, and a personal rank memory that has a rank value set exclusively for registered specific scribes, A standard rank memory with a rank value set for an unspecified number of scribes is provided, and prior to character reading, a personal code is input by key operation or information written on an input form, and the personal code is stored. Character selection is performed by switching between referring to the personal rank memory for registered specific scribes and the standard rank memory for unregistered scribes based on the above. Character selection method.