JPS58222384A - Discriminating system of font - Google Patents

Abstract

PURPOSE:To discriminate the font of a character pattern, by extracting the profile position viewed from the lateral direction of a character pattern picked up one by one character and extracting a characteristic parameter relating to the state of change of the profile position. CONSTITUTION:A reading section 100 scans optically an orginal (not shown) and decomposes the contrast information on the original into pictures for reading them and transmits them as analog picture signals. The picture signal is converted into binary black-and-white picture data at a pre-processing section 102 and noise elimination is processed also. A character pickup section 104 to which the binary picture data is inputted picks up an individual character pattern from the inputted picture data. The character pattern is transmitted to a character recognizing section 112 and a font recognizing section 105. The font recognizing section 105 consists of a left profile extracting section 106, an italic number extracting section 108 and a discriminating section 110 so as to recognize the font (writing style) of the picked up character pattern.

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field The present invention relates to a technology for identifying a font (font) of a character pattern in an optical character reading device or the like.

Prior art When performing character recognition of a character pattern read from a document using an optical reading device, it is advantageous to know the font of the character pattern in advance in terms of improving the recognition rate and increasing the efficiency of the recognition process. be.

If the font of all characters on the manuscript is the same.

Before entering the document into an optical character reader, you can achieve the above benefits by manually specifying the font, or by writing font specification information in a specific position on the document and having the device read it. It is possible to obtain. However, in the case of a manuscript that uses a mixture of different fonts, the above method cannot be practically adopted. In the case of such manuscripts, the above benefits cannot be obtained unless the font can be identified on the basis of individual characters or groups of characters such as words. However, an excellent font identification technique for achieving such a purpose has not been known so far.

OBJECTS Accordingly, an object of the present invention is to provide a font identification method that can identify fonts in units of individual characters or groups of characters such as words. However, in the font identification method of the present invention, the unit of font identification is essentially arbitrary, and it goes without saying that it is also possible to identify the font of the entire document1. The font is determined by paying attention to the characteristics of the seen outline.

Therefore, the font identification method according to the present invention extracts the contour position of a character pattern cut out one character at a time when viewed from the horizontal direction, extracts the characteristic parameters related to the state of change of this contour position, and extracts the character This method is characterized by determining the font of a pattern.

When identifying fonts in units of words, lines, etc., the feature parameters extracted from 1 to 1 are statistically processed for a plurality of character patterns, and the font is determined using the results.

Embodiment FIG. 1 is a block diagram of an optical character reading device to which the present invention is applied.

The reading unit 100 is composed of a known CCD scanner or the like, and optically distorts a document (not shown) at a constant rate, decomposes the grayscale information on the document into pixels, captures 1 νt, and sends it out as an analog image signal. do. This analog image signal is converted into black and white binary image data by the pre-processing section 102, and is also subjected to noise removal processing.

The character cutting unit 102 to which this binary rotational image data is input cuts out individual character patterns from the input image data using a technique such as a well-known projection method. An example of the character pattern cut out here is shown in FIG. This example is an rOJ pattern, with the pattern on the left in italics and the pattern on the right in roman. Such a cutout pattern is sent to the character recognition section 112 and the font identification section 105.

The font identification unit 105 identifies the cut out character pattern □.
``''11111... This is the part that identifies the phono (IT). In this example, there are two fonts: italic and roman.
The font identification unit 105 includes a left outline extraction unit 106, an italic number extraction unit 108, and a determination unit 11.
0 is also configured.

The left contour extraction section 106 scans each horizontal line of the input character pattern from left to right, extracts the position of the first detected black pixel as the left contour position on that horizontal line, and uses it as the italic number extraction section. Send to 108.

The italic number extraction unit 108 stores the left contour position sent from the left contour extraction unit 106 in an internal memory (not shown), and extracts the italic number, which is a characteristic parameter related to the state of change of the left contour position. . In other words, the left contour position is sequentially traced from the upper horizontal line to the lower horizontal line, and the difference (a) in the left contour position between each line and the previous line is calculated.
, and the line in front. Execute the same operation from bottom to top. 1. Difference between the left contour position corresponding to a above (d) and distance from the previous change line corresponding to b above (e) Find each.

Next, italic points are detected from these values, and the number of italic points (I) is determined.

Here, the italic point is the difference a (or d) in the contour position.
is +1 or -1 and/or a change line whose distance b (or d) from the previous change line is 3 or more. The italic number 1 of the italic point is +1 if the difference a (or d) in the contour position of the italic point is +1, and -1 if it is -1. However, ■=-1
The italic point indicates a change in the left contour position (upward stroke to the right) that is typical of italics, and the italic point ■2+1 negates this change. In other words, if it is an italic character pattern, many italic points with ■ = -1 will be found, and the total value of the italic numbers of the valley italic points is one index for determining whether or not the character pattern is italic. It can be used as Therefore, the italic number extracting unit 108 calculates the total number Zl of the italic numbers of each italic point obtained by the procedure described above, and sends it to the determining unit 110.

As mentioned above, the reason why only transition lines with small differences in outline positions (a, d) are extracted as italic points is to eliminate the influence of diagonal strobes that originally exist in curved parts of character patterns. .

Here, let us consider as an example the italic pattern of the letters rOJ shown in FIG. FIG. 3 is an enlarged version of this italic pattern, and the line numbers and horizontal positions are shown as shown.

The results of the processing for this character pattern are shown in a list in FIG.

That is, when examined from top to bottom, ya 9, ba 14, ba η
Each line is extracted as an italic point (downward italic point), and the number of italics for each is as shown in column C'. Also, when examined from bottom to top, A5, B8, A
Each of the 13 lines is an italic point (upward italic point)
The number of italics is as shown in column f. Thus, for this example pattern, 1. The total value Zl of italic numbers is -4.

Now, the determining unit 110 determines whether the character pattern is an italic type or a Roman type by comparing the italic number total value ZIk provided by the italic number extracting unit 108 with an appropriate determination threshold value. Through this determination, for example, the font of the character pattern y (ZI=-4) in FIG. 3 can be identified as italic.

The character recognition unit 112 performs character recognition on the character pattern cut out by the character cutout unit 104, and at this time, it is possible to narrow down the fonts to be recognized according to the results of the font identification described above. Therefore, the recognition rate can be improved and the recognition processing time can be shortened compared to when the font is unknown.

Now, in the explanation so far, font identification has been performed in units of individual characters, but it is also possible to identify fonts in units of character strings such as words or lines. In this case, the determination unit 110 statistically processes the total number of italics obtained in the above-described procedure for one or more character patterns included in the identification unit, and determines the font by comparing the result with the determination threshold. judge.

For example, consider a case where the font of the character string (word) "January" is to be identified between italic and roman fonts. In this case, the total number of italics of each character pattern in the character string of each font and the total value are as shown in Table 1 below.

As shown in the upper table of Table 1, the sum value of italic type is negative (-9), while that of Roman type is positive (+2), and the font can be clearly distinguished.

Note that when performing such statistical processing, it is also possible to exclude from the statistics character patterns with shapes that are likely to cause erroneous determinations, such as the character ryJ.

This can be done, for example, by taking statistics excluding the total italic number of +1 or more. In the upper January character string, the Zl value for the pattern of the character "y" is excluded, and the total value in that case is the value shown in parentheses in the total column of Table 1.

Although one embodiment has been described in detail above, each block in FIG. Not specifically shown.Needless to say, each block can be realized by hardware or by a combination of hardware and software.In addition, each block is shown divided based on function. However, it may be physically realized partially or entirely in one block.

Effects As explained above, according to the present invention, fonts can be identified in arbitrary units such as individual character patterns, character strings, or lines. Therefore, it is possible to identify fonts even for documents containing different types of fonts. . Therefore, if the present invention is applied to, for example, an optical character reading device, it is possible to improve the recognition rate and recognition speed.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an optical character reading device that is an embodiment of the present invention, FIG. 2 is a diagram showing examples of italic and roman character patterns, and FIG. 3 is an example of italic character patterns. FIG. 4, which is an enlarged view, is a diagram showing the contents of the font identification process for the character pattern of FIG. 3. 104... Character cutting section, 105... Font identification section, 106... Left outline extraction section, 108... Italic number extraction section, 110... Judgment section. 4, ok,, oh, +, 7 oh)゛. □ 1-1

Claims (2)

[Claims] (1) Extract the outline position seen from the horizontal direction of the character pattern cut out one character at a time, extract the feature parameters related to the state of change of the outline position, and determine the font of the character pattern from the extracted feature parameters. A font identification method characterized by: (2) The font identification method according to claim 1, wherein the font is determined based on the statistical processing of feature parameters extracted from a plurality of character patterns without performing statistical processing.