JPS6182284A - Stroke extraction method - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to a method for extracting strokes of a character pattern in a single character recognition device.

[Prior art]

The stroke extraction method used in conventional character recognition devices is to encode character patterns by direction, then perform decompression processing to extract strokes by direction, and separate connected strokes using background information. ing. However, in such conventional methods, even if the character patterns are extremely similar, the number of extracted strokes may differ, making the feature values unstable, which increases the dictionary capacity and reduces the recognition rate. There is also a problem that the recognition speed is lowered.

Such an example will be explained with reference to FIG. This figure is (a
) is the direction code LL obtained by direction-specific coding.
1, l pattern. In the conventional method, four strokes are extracted from this pattern as shown in (b). In (b), 111 II to 4'' are stroke identification numbers, respectively. (c) is a putter with direction code LL I II obtained by the direction code, and according to the conventional method, from this pattern ( Three strokes are extracted as shown in d).

In this way, the pattern (C) differs from the pattern (a) only in that the two pixels circled in the center are white, but from the pattern (a), there are four pixels. strokes are extracted, whereas only three strokes are extracted from the (c) pattern.

〔the purpose〕

An object of the present invention is to provide a stroke extraction method that solves the above-mentioned problems.

〔Example〕

FIG. 2 is a schematic block diagram of a character recognition device according to the present invention. The input document is converted into binary image data by the photoelectric conversion unit 100. The preprocessing unit 101 cuts out characters from the photoelectrically converted image data, smooths them,
Performs processing such as normalization. The feature extraction unit 102 extracts a feature amount from the character pattern cut out by the preprocessing unit 101 and sends the feature amount to the search unit 103. Search section 10
3 compares the extracted feature amount with the feature amount of known characters registered in advance in the dictionary storage section 104 to perform single character recognition, and sends the recognition result to the output section 105.

The output unit 105 prints, displays, or stores the recognition results.

The configuration of the feature extraction section 102 will be explained with reference to FIG. In this figure, 301 is an overall feature extraction unit that extracts the overall features of a character, such as the center and number of holes, the center of gravity of the character, etc. The other parts are related to stroke extraction, and their configuration is as follows.

The character pattern input from the preprocessing unit 101 is coded by direction by the direction coding unit 302. This coating is performed, for example, by a known method of checking the connection direction of black pixels of a character pattern using a 3×3 mask, and attaching a direction code corresponding to the black pixel according to the connection direction.

The decompression unit 303 decompresses each direction code for the coded pattern as long as the pixel is black in that direction. An example of the pattern after this expansion is shown in FIG. 4(a). In this figure, LL i 11 is the direction code.

The stroke extraction unit 304 extracts strokes from the expanded pattern while checking the connection state in each direction. At this time, for a pattern in which strokes are connected as shown in FIG. 4 (Lt), the strokes are separated using background information. This process itself may be the same as the conventional one. Further, when a stroke connection is detected, the connection afragment Fi in the stroke direction is turned on.

In the case of the pattern shown in Fig. 4(a), as shown in Fig. 4(b), 4
Two strokes are extracted. In this figure, LL
I II to II 4 II are stroke identification numbers.

After extracting strokes, the stroke extraction unit 304 extracts stroke characteristics such as center coordinates, length, and width of each stroke.

When the stroke extraction section 304 finishes stroke extraction, the thinning stroke extraction section 305 sets the connection flag F.
Check the status of i. If the connection flag Fi is in the OFF state, the thinning stroke extraction unit 305 outputs a processing end signal to the stroke selection and integration unit 306 without doing anything. If the connection flag Fi is in the ON state, the thinning stroke extraction unit 305 thins the pattern of each direction code after being expanded by the expansion unit 303. This thinning process is conventionally known. method)
, extract strokes from the thinned pattern, and extract the center coordinates of each stroke.

When this process is completed, a process end signal is output.

In the case of the pattern shown in FIG. 4(a), the thinned stroke extraction unit 305 extracts four strokes as shown in FIG. 4(c).

When the stroke selection and integration unit 306 receives the processing end signal from the thinning stroke extraction unit 305, it checks the state of the connection flag Fi. When the connection flag Fi is in the OFF state, the strokes extracted by the stroke extraction section 304 are directly selected as the final strokes, and their stroke characteristics are output to the search section 103.

When the connection flag Fi is in the ON state, the stroke selection integration unit 306 selects the number of strokes Mi extracted by the stroke extraction unit 304 and the thinning stroke extraction unit 305.
The number of strokes Ni extracted by is compared.

If this comparison matches, the stroke selection integration unit 30
6 selects the strokes extracted by the stroke extraction section 304 as the final strokes, and outputs their stroke characteristics to the search section 103.

If the above comparison does not match (M i )N i), the distance between the center coordinates of the strokes extracted by the stroke extraction unit 304 and the strokes extracted by the thinning stroke extraction unit 305 (relative distance) Calculate. Then, the stroke (M
i books), N4 in order from the one with the smallest relative distance
Select only books.

Furthermore, for the remaining unselected strokes, the stroke feature closest to the center coordinates of the selected stroke is calculated. This integrated stroke and the selected stroke are extracted as final strokes, and their stroke characteristics are output to the search unit 103.

Note that the above-described processing by the thinning stroke extraction unit 305 and the stroke selection unit 306 is executed for each direction.

In the case of the pattern shown in FIG. 4('a), four strokes are extracted by both the stroke feature extraction section 304 and the thinning stroker extraction section 305, so each stroke shown in FIG. The stroke characteristics are output to the search unit 103.

FIG. 5 shows an example in which the number of strokes does not match.

(a) is the pattern of the direction code "'1" after encoding and expansion, and (b) is the pattern of strokes extracted by the stroke extraction unit 304.The center coordinates of each stroke are shown in the table. (c) is a stroke pattern extracted by the thinning stroke extraction unit 305, and the center coordinates of each stroke are as shown in the right column of Table 1.

In this case, the stroke Nα1. extracted by the stroke extraction unit 304. Nα4 is selected as the final stroke, but strokes Nα2 and No.
, 3 are integrated into one stroke, and the integrated stroke is finally extracted.

Table 1 In addition, encoding by the direction-based encoding unit 2 and decompression unit 30
3, stroke extraction and separation in the stroke extraction section 30/1, and thinning and stroke extraction processing by the thinning section stroke extraction section 5, as described above. This can be done in a similar manner. The feature of the present invention lies not in such individual processing, but in a series of processing sequences as described above.

〔effect〕

As explained in detail above, according to the stroke extraction method of the present invention, it is possible to prevent fluctuations in the number of strokes due to subtle differences in character patterns and perform stable stroke extraction. reduce capacity,
Furthermore, it is possible to improve the recognition rate and recognition speed.

In addition, since the stroke extraction process after thinning is executed only when a connection of strokes is detected in the non-thinning stroke extraction process, the increase in stroke extraction processing time due to the addition of the thinning stroke extraction process is minimized. can be kept to a minimum.

[Brief explanation of drawings]

FIG. 1 is a pattern diagram for explaining conventional problems, FIG. 2 is a schematic block diagram of a character recognition device according to the present invention,
FIG. 3 is a detailed block diagram of the feature extraction section, and FIGS. 4 and 5 are pattern diagrams showing examples of stroke extraction. 100... Photoelectric conversion section, 101... Pretreatment section,
102... Feature extraction unit, 103... Search unit, 1
04... Dictionary extraction section, 105... Output section, 30
DESCRIPTION OF SYMBOLS 1... Whole feature extraction part, 302... Direction-specific coding part, 303... Expansion part, 304... Stroke extraction part, 305... Thinning stroke extraction part,
306...Stroke selection integration unit. Cleaning of the drawings (no changes to the contents, Figure ,, 1111111111.. ,,,, 11111111... ...,, 11111.,,,, cL) ......444 ...... ,,, 3333... ,,,3.222222222・. -1,,22222222,,. 0. ,． 222>2. ? 22... ,, -2222222222. (C') ,,,1111111111. . -,,1111111117-. ...0111...,. ...11111171... ・,,1111111111. . ...11111... (匈...3RO3... ,,,,33333333.,.,,,3333333333-.10.333"333" 1133...,,,222
．． ,,,,,. ,,,, 222222222... ,,,,22222)22. ,.豐#感-22222ふ'22...-0・2222ふ22722... ,,,1'X11111111. . ...11Tl1111... ...1llll... Procedural amendment (method) 1. Display of the case Patent Application No. 58-247188 Invention 2, Name of M Person who makes the stroke Tsukudade method 3 amendment Relationship to the case Applicant address 1-3-6-8 Nakamagome, Ota-ku, Tokyo! '
F (674) Ricoh Co., Ltd. -4 Agent address: No. 5, Line Department Building 201, 2-38-X2, Yoyogi, Shibuya-ku, Tokyo 8I51 Date of amendment order: March 7, 1980 (Date of dispatch: 1982) (March 4) Revise Figure 1 as shown in the attached sheet. 9. Amendment to list of attached documents 1 copy of drawings Procedure amendment (method) October 28, 1985 Patent Application No. 247188 filed in 1988 2. Invention name stroke extraction method 3. Related Applicant Address 1-3-6 Nakamagome, Ota-ku, Tokyo Name (
674) Ricoh Co., Ltd. - Representative
Hama 1) Hiro 4, Agent 5, Date of amendment order: August 7, 1985 (Shipping date: August 27, 1985)
7. Contents of amendment Section 11, lines 14-15, “, Figures 4 and 5
The description "The figure is a pattern diagram showing an example of stroke extraction" will be deleted. that's all

Claims (1)

[Claims] (1) The first step is to attach a direction code to each black pixel of the character pattern according to the connection direction, and then expand each direction code, extract strokes for each direction code, and check whether the strokes are connected. and a second step of extracting the strokes of each direction code after thinning the pattern of each direction code after the expansion when a stroke connection is detected in the first step; If a stroke connection is not detected, the stroke extracted in the first stage is extracted as the final stroke, and if a stroke connection is detected in the first stage, the stroke extracted in the first stage and the second stage are extracted. If the number of strokes extracted in the first stage is the same, the stroke extracted in the first stage is extracted as the final stroke, and if the number of strokes does not match, the strokes extracted in the first and second stages are The relative distance is calculated, and from among the strokes extracted in the first stage, the same number of strokes as the number of strokes extracted in the second stage are selected in descending order of relative distance, and the remaining strokes that are not selected are selected. A stroke extraction method comprising a third step of selectively integrating strokes with small relative distances to extract a final stroke.