JPS61168080A - Character recognizing device - Google Patents

Abstract

PURPOSE:To improve the recognition rate of inclined characters, by correcting the inclination of characters before performing character recognition. CONSTITUTION:A scanner 10 outputs picture signal of an original and a pre- processing section 12 performs binary coding process and character cutting-off process on the picture signals. Cut-off character picture data are accumulated in a character picture data memory 14 and read out by an inclination correcting section 16, where inclination correction is performed to the data by angles of 0 deg., -15 deg., -30 deg., and -40 deg.. A slanting shadow width detecting section 18 extracts vertical shadows of the character picture data which are inputted from the inclination correcting section 16 after inclination correction of each angle and detects each shadow widths then compares the detected shadow with each other and selects the nimimum one. The correcting angle of the minimum shadow width is set in the correcting section 16 and the character picture data corrected with the correcting angle are outputted to a recognizing section 22 where character recognition is performed with special attention being paid on the directional property of character picture boundaries.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to a single character recognition device, and more particularly to a character recognition device that performs character recognition after correcting the inclination of a character.

[Prior art]

As an effective method for recognizing handwritten characters, "Handwritten Kanji/Hiragana Recognition Using Weighted Directional Index Histogram and Pseudo-Bayesian Discrimination Method" by Tomoo Harada et al. (IEICE Technical Report PRL-8)
2-79 and IEICE Technical Report PRL-83-68), there is a method for determining the directionality of the boundary of a single character image. However, in this method, there is a problem in that the influence of variations in the slope of one character line is large, and the recognition rate of handwritten characters that are tilted diagonally upward to the right or downward to the right is poor. 'For example, IIN'' shown in Figure 3(a) is
The direction component is close to 'W' shown in , and it is mistakenly recognized as il Wj#.

[Eye□target]

The present invention aims to improve the recognition rate of tilted characters by correcting the tilt of the characters prior to character recognition.

〔composition〕

In the present invention, the projection of one character image is obtained at a plurality of angles, the angle with the smallest projection width is selected, and the inclination of the character image is corrected at the selected angle before character recognition is performed.

Hereinafter, one embodiment of the present invention will be described with reference to one drawing.

〔Example〕

FIG. 1 shows a schematic block diagram of an optical character recognition device according to an embodiment of the present invention. In this figure, scanner 1
0 outputs an image signal of the original by scanning and photoelectrically converting the original. The preprocessing unit 12 performs preprocessing such as binarization and character extraction of the image signal, and the image data of the extracted characters is stored in the character image memory 14.

The tilt correction unit 16 reads the character image data from the character image memory 14, and reads the character image data at 0°, -15° (- sign means a counterclockwise angle), and -30°.

Tilt correction is performed for each angle of =45°. Incidentally, such tilt correction processing can be easily executed by the method described in Japanese Patent Laid-Open No. 58-87652 entitled "Image Processing Apparatus". The projection width detection section 18 extracts the vertical projection of the character image data after the tilt correction for each angle input from the tilt correction section 16, and detects the width of each projection. Correction control section 20
Now, the projection widths at each correction angle detected by the projection detection section 18 are compared, and the minimum projection width is selected. Then, the correction angle of the minimum projection width is set in the skew correction section 16, and the character image data whose skew has been corrected at that angle is input from the skew correction section 16 to the recognition section 22. The recognition unit 22 performs character recognition on the character image data input from the 1ifl correction unit 16, focusing on, for example, the directionality of the character image boundary as described above.

For example, in the case of the character "N" in FIG. 3(a), the projection of each angle is as shown in FIG. In this N.

30 is a projection of a character image with 0° correction (no correction), 32 is a projection of a character image with -15° correction, 34 is a projection of a character image with -30° correction, and 36 is a projection of a -45° corrected image. . Projection width WllTWztlW for each correction angle
Among inlW4+t, W31. is the minimum. Therefore, the correction control unit 22 sets the correction angle to -30° in the tilt correction unit 16. In this case, the corrected °'N" input to the recognition unit 22 is as shown in FIG. 4(a), and is sufficiently close to the dictionary standard pattern "N" shown in FIG. 4(b). Dictionary standard pattern ゛l W shown in figure (c)
It is clearly distinguished from p''.

In this way, since the tilt of the character image is appropriately corrected prior to character recognition, the recognition rate of tilted handwritten characters is greatly improved. Also. The determination of the tilt correction angle is also simple, the device configuration is not so complicated, and a decrease in processing speed is not a problem.

In the first embodiment, the tilt correction was performed for each character, but in the second embodiment, the tilt correction is performed only for necessary characters according to the result of character recognition. For more information,
The explanation will be given with reference to the block diagram of FIG. 1 again.

The correction control unit 20 normally sets 0° as the correction angle in the tilt correction unit 16. In other words, 1 usually.

The character image data read out from the character image memory 14 is directly sent to the recognition unit 22 via the skew correction unit 16, where the character is recognized. The recognition unit 22 selects a first candidate character having a minimum distance fid from the input character.

Then select the second candidate character with the next smallest distance lid.

If d, <DTH and (d, -d,)>KTH, then (D
TH and KTH are constants), and the first candidate character is output as a recognition result. If the above conditions are not met, the recognition unit 22
The correction control unit 20 is instructed to perform tilt correction.

When the correction control unit 20 receives the tilt correction instruction.

As in the case of the first embodiment, the inclination correction unit 16
The inclination is corrected at each angle of -30° and -45°, and the projection detection unit 18 is caused to detect the projected width of the corrected character image at each angle. Then, the correction control unit 20 selects the smallest of each projection width, sets the corresponding correction angle in the skew correction unit 16, and transfers the character image data corrected at that angle from the skew correction unit 16 to the recognition unit. 22 and have it recognized.

The recognition unit 22 outputs the first candidate character for the character image after the tilt correction as the final recognition result.

In this embodiment, since the number of tilt corrections is reduced, the processing speed can be increased accordingly.

〔effect〕

As explained above, according to the present invention, the inclination of a character image is corrected prior to character recognition.

The recognition rate of handwritten characters with large inclinations can be improved.

In addition, since the tilt correction angle is appropriately determined using an east line method, the device configuration does not become so complicated and a decrease in processing speed does not become a problem.

[Brief explanation of drawings]

Fig. 1 is a schematic block diagram showing an example of an optical character recognition device according to the present invention, Fig. 2 is an illustration of projections at different angles and their widths, and Fig. 3 is an illustration of misrecognition due to character inclination. FIG. 4 is a diagram showing a character pattern for explaining the effect of tilt correction. 10...Scanner, 12...Preprocessing section. 14... Character image memory, 16... Tilt correction section, 18... Projection detection section, 20... Correction control section. 22... Recognition section. (α) (b) Ritsu) (shi) (C)

Claims (1)

[Claims] (1) Means for determining the projection of a character image at multiple angles and detecting the width of each projection, means for selecting the angle with the minimum width of the projection detected by this means, and the angle selected by this means. A character recognition device comprising: means for correcting the inclination of a character image and inputting the corrected character image to a character recognition unit.