JPS6033678A - Underline detecting method - Google Patents

Abstract

PURPOSE:To detect accurately an underline connected to characters by recognizing characters after masking a unit of character image data and an underline candidate part by a horizontal projection and deciding the underline from the result of character recognition. CONSTITUTION:A character cutting-out underline detecting part 4 cuts out each character image data within an image memory part 2 for each break of a vertical projection detected by a vertical projection detecting part 3 and then gives a horizontal scanning to the lowest column from the highest column of the cut- out image data to detect the top and bottom parts of the image. Then a horizontal scanning is carried out from the bottom part to the top part to obtain a horizontal projection. The presence of an underline candidate is decided if a horizontal projection corresponding to the underline length exists at the underline position. Then the underline candidate is masked when the existence of just one horizontal projection is decided. Then characters are recognized at a character recognizing part 6. Based on the result of this recognition, an underline is detected.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to an underline detection method in a character recognition device.

[Prior art]

In order to recognize underlined characters in a character recognition device, mark the underlined part from the character image data cut out in units of characters.
, I/,゛Moe+i+-11z(1'+ +-out1.-
It is necessary to detect t+, 7%7 J/--'r-/n.

In the conventional underline detection method, a horizontal white run of a predetermined length or longer continues vertically in character image data, followed by a black run of a predetermined length or longer, and below that a white run of a predetermined length or longer. If it continues, it is determined that there is an underline (Japanese Patent Laid-Open No. 56-500905). However, with this method, underlines connected to characters cannot be detected correctly, which may cause recognition errors.

〔the purpose〕

An object of the present invention is to provide an underline detection method that can accurately detect underlines connected to characters.

Another object of the present invention is to provide an underline detection method that can contribute to improving the recognition rate of underlined characters.

〔Example〕

FIG. 1 is a schematic block diagram of a character recognition device according to the present invention. Image data input from a scanner (not shown) is input to a noise removal/line cutting section l,
Isolated point removal and line cutting are performed [row]], and the image data for one line that has been cut out is stored in the image memory section 2. At this time, the vertical projection detection section 3 detects the vertical projection of one row of image data. Incidentally, line cutting is performed by a known projection method, etc., and the image data including the underline portion is cut out. Ru.

When line cutting is finished, the character cutting/underline detecting unit 4 cuts out individual character image data in the image memory unit 2 for each vertical projection break date detected by the vertical projection detecting unit 3, and then cuts out the individual character image data in the image memory unit 2. Processing according to flowchart 1- in the figure is performed.

First, the character image data is horizontally scanned from the top row to the bottom row, and the first row in which a black pixel is found is detected as the top of the image, and the last row in which a black pixel is found is detected as the bottom of the -r image (G>). Next, horizontally scan the character image data from the bottom to the top to obtain the horizontal projection (+21
).

If there is a horizontal projection of a value (the expected number of pixels) corresponding to the length of the underline at the underline position, it is determined that an underline candidate exists (■).

If it is determined that there is no underline candidate, the character image data is sent as is from the image memory section 2 through the masking section 5 to the character recognition device 6 for final character recognition ((■).The recognition results are sent to the host (not shown). (not sent).

If there is an underline candidate, water) V-projection is 1
Determine whether there are only one (■). If the characters and underline are separated as shown in Figure 3, the judgment result is N.
It becomes O. If the characters and underlines are connected as shown in Figure 4 or Figure 5, the judgment result is YES.
bawl.

If the determination in step (■) is No, the underline candidate is detected as an underline, and the character image data is sent from the image memory 2 to the character recognition unit 6 for final character recognition. Mask the underline candidate part of the character image data,
Also, add an underline code (0+@). The recognized character code is sent to the host with an underscore code 1- added thereto.

Step (If the top determination result is YES, the character image data is sent to the character recognition unit 6 with the underline candidate portion masked, and provisional character recognition is performed (■, ■).The result of provisional character recognition is υ in the out/underline detection section 4,
available.

It is determined whether the character recognized by the temporary character recognition is rr 9 n corresponding to the pattern in FIG. 5 ((S)).

If YES, "Resheku l--9" = 1-1-4 characters a is sent to the host via the intelligence unit 6, and the host performs recheque 1~processing (■)). Such character image data is 9″′ with an underline, “g” with an underline, or 11
Since it is not possible to distinguish whether g++ has an underline or not, rejiku 1~ processing is performed. Relayen 1-processing on the host side is (1) simply processing it as a character to regiator 1, (11) if the preceding and following characters do not have an underline, they are marked with 'g゛', and the underlined characters are If so, display the symbol 1~ and -1r/-? l
t -JJ i-) X primary if; ly 'J>
#'l R-I'm very talented.

If the judgment result in step ■ is NO, the characters recognized by temporary character recognition are "y"' (Figure 4), "l" (vertical bar), 11
Determine whether there is only one J of p Hllll, ++ (
(11). Each of these characters can be easily distinguished even in character image data with the underlined part masked, so
There is no problem even if the result of temporary character recognition is determined, and underline candidates may be detected as underlines. Therefore, if the result of judgment &11 is YES, the character code obtained by temporary character recognition is added with an underline code and 1
The character recognition unit 6 sends it to the host (■).

If the determination result in step (2) is No, the character image data is a capital letter "E", "z", etc., and the underline candidate is not an underline.

Therefore, the character image data is sent as is to the character recognition unit 6 without masking the underline 1 complement part, and fixed character recognition is performed (@).

In addition, in the provisional character recognition in step
I I ++, 11 PIl, If yll, I
Since it is only necessary to know which of the image data of Z 11 matches, it is possible to use a simple pattern comparison means such as a number change, in addition to the character t and identification part 6, to perform the process.

〔effect〕

As described in detail above, the present invention detects underline candidates by horizontal projection, masks character image data with underline candidates and one horizontal projection, and then performs character recognition. Based on the result, it is determined whether the underline candidate is a true 7-dahline or not, so it is possible to accurately detect underlines connected to characters, and it also prevents literary recognition errors due to false detection of underlines; plagiarism recognition. , towards i4i'
It has the effect of being able to do two things.

[Brief explanation of the drawing]

FIG. 1 is a schematic block diagram of a character recognition device according to the present invention;
Figure 2 is a flow diagram of underline prevention processing, Figure 3 is a diagram showing an example of an underline and character pattern, and Figures 4 and 5 are patterns in which underlines and characters are connected. It is a figure showing an example. 1... Noise removal/truncation peak part, 2... Image memory part, 3... Vertical projection detection part, 4... Character extraction/underline detection part,' 5... Masking part, 6... Character Recognition part. Representative Patent Attorney Makoto Suzuki・°゛、c) Figure 1 Figure 2

Claims (1)

[Claims] (1) Determine whether an underline candidate exists based on the horizontal projection of character image data, and for character image data that has an underline candidate and one horizontal projection, perform character recognition by masking the underline candidate part. An underline detection method characterized by determining whether an underline candidate is an underline based on the result.