JPS6072088A - Method for selecting dictionary set of character form corresponding to character form used in original - Google Patents

Abstract

PURPOSE:To improve an accuracy for recognizing a character form by selecting a dictionary of a character form corresponding to a character form used in an original, from in plural dictionary sets prepared at every character of different character forms. CONSTITUTION:In a dictionary retrieving part DRD, with regard to each character which has been read from an original, a distance arithmetic of a feature parameter from a feature extracting part CAD, and a dictionary set is executed at every dictionary set 1-3. In case a difference (size of a distance) of the dictionary and the feature parameter is outside of the allowable range, the character is regarded as an unreadable character. The number of unreadable characters is counted by counters 6-8 provided on every dictionary set. In case a dictionary set in which the number of unreadable characters is the smallest is only one, its dictionary set is selected, and in case two or more dictionary sets in which the number of unreadable characters is the smallest exist, the dictionary set in which a total value of the distance is the smallest in those dictionary sets is selected.

Description

[Detailed Description of the Invention] (Technical Field) The present invention provides optical character reading! A. This paper relates to a technique used in connection with the identification of fonts used in the recognition of characters (characters, cross-cut characters) of various fonts.

(Prior Art) Conventional methods used to identify the fonts of many types of characters in optical character reading devices include, for example,
As disclosed in Japanese Patent Publication No. 5G 44455, when a reading operation is performed using a plurality of dictionary cellars, the output result from the dictionary cellar I. which generates the least number of unreadable characters is selected. Recruitment or, for example,
The distance t between the extracted feature parameters and the dictionary of fonts
However, in the case of song titles using the conventional method described above, it is difficult to identify misidentified characters, and there is overlap between parameters for the same character type in different fonts. In some cases, there is a problem in that the fonts are identified by a cone, and in the latter case in the conventional method described above, there is no obvious overlap between the parameters of the same character type of different fonts (for example, when there is no overlap between the parameters of the same character type in different fonts (Purpose) The present invention provides a dictionary set of 7 fonts corresponding to the fonts of the manuscript, which can improve the recognition rate of characters with many types of fonts. It is our objective to provide a method for selection so that the problems with the conventional methods described above can be solved.

(t-f composition) Hereinafter, appended A.
(\Dobun 4 How to select the corresponding font in the dictionary,
4r That is, for each character with a different font, select the ``゛, font, and sentence, I Lt,: ``, ``, font, and sentence'' used at cost from a dictionary set of numbers. In a method of The distance between the feature parameter and the dictionary is calculated, and characters whose distance is larger than a predetermined value are considered unreadable characters, and the number of characters is counted for each dictionary cellar 1-, and the number of characters is counted for each dictionary cellar 1-. Calculate the total value of the distances calculated in 7, and if there is only one dictionary set with the least number of unreadable characters, select that dictionary set,
In addition, if there are two or more dictionary sets with the minimum number of unreadable characters, the one with the minimum sum R1 value of the visible part is selected from among those dictionary sets. We will give a detailed explanation of how to select a dictionary set of corresponding fonts.

FIG. 1 is a block diagram schematically showing the configuration of an example of an optical reading nail to which the method of selecting a dictionary set of fonts corresponding to the font of a manuscript according to the present invention is applied. F'EC is an optical character reading section (photoelectric conversion section), FJJD is a preprocessing section, CAD is a feature extraction section, L)-RD is a dictionary IF search section, and the second
The figure shows a block diagram of a part of an embodiment of the dictionary search unit DRD to which the method of selecting a dictionary set of fonts corresponding to the font of a manuscript according to the present invention is applied, and further, FIG. 3 shows a flowchart. ing.

In FIG. 2, 1 to 3 are dictionary sets with different fonts, 4 is a search control section, 5 is a distance calculation section,
6 to 8 are distance adders, and 9 to 11 are counters for counting the number of unreadable characters.
And 8! The number M) 9 handles data related to the dictionary set 1, and the pressure part addition W7 and the 1 digit unit 10 handle the data regarding the dictionary set 1.
It is assumed that the distance adder 8 and the counter 11 handle data belonging to the dictionary set 3 (in the example shown in the second figure, the font type is 37J). The case is shown here, but the Tff type of the font is arbitrary and °C is also fine, and the \'TX type of the font is 3g of M.
6 also uses M pieces of each of the external adders and multimeters mentioned above! :J: l, and each one dictionary set I- in one dictionary set of one tome needs to be associated with one distance adder and one counter, so there are five entries).

Now, the characters on the original are one optical character reading section (photoelectric conversion section).
The image signal is converted into an image signal by PEC, and the image signal is subjected to predetermined signal processing by the preprocessing unit FDD, feature extraction unit CAD, etc., and the feature parameters of the characters in the document are stored in the dictionary search unit DRD. given to CAD.

The dictionary search unit DRD calculates the distance between the feature parameters given to each character from the feature extraction unit CAD and the dictionary set for each character in a predetermined number N of characters read from the manuscript to each dictionary cellar 1. Do this every 1 to 3 times.

For example, the distance calculation described above is performed using Rji for the distance 11R.
=Σw l dkj−Pki l −−・・・・・・(
1) k=1 This can be done by considering a function such as the above equation (1). In equation (1), Rji is the distance history between the dictionary vector dj and the feature vector Pi, W is the weighting function, dkj is the component (parameter) of the dictionary vector dj, and P
hi is the h component (parameter) of the feature vector Pi.

When calculating the distance Qi, the allowable range of the difference between the dictionary and the feature parameter (the size of the distance R1) is given to the parameter σ.
It is predetermined by the weighting function W described above, and if the difference (size of distance 1 get t) between the dictionary and the feature parameters described above is outside the tolerance range described above, those outside the tolerance range are read. The characters are treated as unreadable characters and excluded from the calculation of distance m, and the number of unreadable characters is counted by a counter provided individually for each dictionary set.

The calculation results for each dictionary set prepared for each font for a predetermined number N of characters are calculated by a distance adder provided individually for each dictionary set. -C is added.

As mentioned above, the dictionary vector and the feature vector are
'l? The ni1 operation is performed, and the feature Betator 1 is recognized as the character code of the dictionary with the minimum distance in Lji,
be recognized.

In other words, it is assumed that there are three types of dictionary sets prepared for three types of fonts, as shown in the example shown in FFS2, and that there are a predetermined number N of dictionaries in the manuscript in order to recognize the fonts. (for example, the N characters following the first character in the manuscript, or for example, the characters for one line in the manuscript) are used, then the above N characters &yo, the above 3 Distance calculations are performed between each type of dictionary cellar 1-.

Then, for each dictionary set, the number nu of unreadable characters is counted by counters 6-8, and the distance of the first candidate for each character is added to distance adders 9-11. Once distance addition has been completed for all of the predetermined number N of characters.

After comparing the counts in the counters 6 to 8, the result is 8! If there is one counter with the smallest value, the dictionary set corresponding to that counter is selected.

In addition, as a result of comparing the count values of the nl counters 6 to 8 described above, it was found that there are two or more counters with the minimum count value.
If the result is 2.

A dictionary set in which the total distance value LRt shown by the following equation does not have a minimum value and corresponds to the distance adder is selected.

FIG. 3 is a flowchart showing how to select the dictionary set described above.

(Effect) As is clear from the above explanation in detail 3.11, the method of creating a dictionary set of fonts corresponding to the fonts of one Ja Ming manuscript is j''!4i; Select a dictionary with a font that corresponds to the font used in the manuscript from among multiple dictionary sets prepared for each character, and then For each word Zset+-fT in cella 1~, calculate the distance between the character parameter and the dictionary using !l' for each of the predetermined number of characters in the manuscript,
Characters whose distance l1ilc is larger than a predetermined value of 4 are treated as unreadable characters, and the function is calculated for each dictionary set by n (and the sum of the distances calculated for each dictionary set is 81). If the number of unreadable characters or the minimum ffF document set is one, then the i
'i Hana j (Hiso 1- is removed from Tosugi, and the unreadable characters II^1 numerical power i, %1. Small 3?, '7: Set is 2
If there are more than -C these words 11; b , ,
.

Since the method selects the font of the font that corresponds to the font of the manuscript by selecting the smallest distance within 1. The accuracy of font recognition can be greatly improved by comparing the numbers.

[Brief explanation of drawings]

Eleventh? i1 is a block diagram illustrating the 4Q jl &4'ff configuration of the optical reading device occupied by the method of selecting the subset of fonts corresponding to the font of a document according to the present invention, f('j 2 lpl is a method for selecting a dictionary set of fonts corresponding to the fonts of the manuscript of the present invention. is flowchart 1. P E C... optical character reading section (photoelectric conversion section), F
1] D...Preprocessing section, CAD...Feature extraction section, DR
D... Dictionary search unit, 1-3... Dictionary cellar 1 of font, 4... Kensu control unit, 5... Pressure part calculation unit, 6-8...
・Distance adder, 9 to 11... Counter for the number of unreadable characters, 1 'm

Claims (1)

[Claims] A method of selecting a dictionary of fonts corresponding to the font used in a manuscript from among a plurality of dictionary sets prepared for each of the different fonts, For each set of if books of the plurality of dictionary sets 1~, for each of the predetermined number of characters in the manuscript, the 1st parameter and the 1♂Ij l' of the τr book are calculated. 71j operation, the above y+x
Characters where m is larger than -F are treated as unreadable characters, and the number of such characters is added to each dictionary set 4if as a (along with the number, the sum of the distances calculated for each dictionary set 1-i σ is 31 values). If there is only one dictionary set with the least number of unreadable characters, select that dictionary set, and select the dictionary set with the least number of unreadable characters: i'!
If there are two or more ``motu 1~'', select the dictionary set of the font corresponding to the font of the manuscript. how to