JPH02135586A - Optical character reader - Google Patents

Abstract

PURPOSE:To recognize both dark character and blurred character by providing first and second means, which respectively binarize signals with a threshold following the white level of a background and the threshold based on the blackest mesh of the circumferential picture element, providing the synthetic deciding parts of a character recognition result corresponding to each binarization in parallel, and parallel- processing the plural binarizing systems. CONSTITUTION:A preprocessing part 2 consists of a first binarizing means 9, which binarizes a quantizing electric signal outputted from a scanner 1 by the threshold following the white level of the background of a paper surface, a second binarizing means 10, which binarizes the signal by the threshold calculated based on the blackest mesh of the circumferential picture element, etc. A character recognizing part 3 provides a recognition control part 17, which matches a standard character pattern stored into a dictionary memory 16 to the picture pattern of each input character from the preprocessing part 2 and recognizes the input character, and further a synthetic deciding part 4 is provided in parallel on an output stage. The synthetic deciding part 4 decides the character recognition result by means of the binarization based on majority decision. Thus, the dark character and blurred character can be more correctly recognized.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention obtains a character image pattern written on paper such as a form by using optical scanning and photoelectric conversion techniques, and also obtains a character image pattern written on a paper surface such as a form, and The present invention relates to an optical character reader (OCR) that recognizes characters.

[Conventional technology]

-i, in this type of optical character reading device, in order to quantize the density value of input image data into binary values of black and white, it is important to determine a threshold value for binarization.

Methods for determining this threshold value can be roughly divided into a fixed threshold method and a floating binarization method.

If the density values of text and background are known and change little from input image to input image, a fixed threshold method in which the threshold is determined empirically in advance is simpler and more practical.

On the other hand, when input conditions such as paper and characters change, it is necessary to determine the threshold value according to the input image. For this reason, the floating binarization method is relatively often used for reading handwritten characters on forms.

In this floating binarization method, there are several methods for setting the level. Among these, most of them are the white level tracking type threshold method, which performs binarization using a threshold that follows the white level of the background of the form, or the black level tracking type variable threshold method, which performs binarization using a threshold that follows the black level of the surrounding mesh. Either is used.

[Problem to be solved by the invention]

However, in the white level tracking type threshold method in the above-mentioned conventional example, the threshold value is determined based on the white level of the background of the form, resulting in the problem that, for example, dark characters are crushed.

Sometimes, the thin parts of the blurred letters would disappear. On the other hand, when performing binarization using the conventional black level tracking type variable threshold method described above, the threshold value is determined by following the black level of the surrounding mesh, so stains, creases, etc. in the white part of the form, etc. The white level tracking type threshold method had the disadvantage of picking up noise that was not picked up.

Therefore, in either case, misreading occurs frequently, the corresponding standard character is not found, or two or more standard characters are selected as candidate characters and rejected as unreadable.

[Purpose of the invention]

SUMMARY OF THE INVENTION An object of the present invention is to provide an optical character reading device that can improve the disadvantages of the conventional example and, in particular, can more accurately recognize both dark characters and blurred characters.

[Means for Solving the Problems] The present invention includes a photoelectric conversion unit that photoelectrically converts reflected light obtained by optically scanning characters on a paper surface, and binarization of an output signal of this photoelectric conversion unit. and a character recognition section that compares and judges an input character pattern obtained based on an output signal from the preprocessing section with a known standard character pattern to recognize the character. In particular, the preprocessing section includes a first binarization means that binarizes at least the quantized electrical signal output from the photoelectric conversion section using a threshold value that follows the white level of the background of the paper, and a second binarization means that binarizes using a threshold calculated based on the blackest mesh, and a character recognition unit that includes a comprehensive judgment unit that comprehensively judges the result of character recognition corresponding to each binarization. The structure is such that it also has a This aims to achieve the above-mentioned objective.

[Embodiments of the invention]

An embodiment of the present invention will be described below with reference to FIG.

The embodiment shown in FIG. 1 includes a scanner 1 as a photoelectric conversion unit that photoelectrically converts reflected light obtained by optically scanning characters on a paper surface such as a form, and a binarization of the output signal of this scanner 1. a preprocessing unit 2 that performs the above operations, and a character recognition unit 3 that recognizes the character by comparing (matching) and determining the input character pattern obtained based on the output signal from the preprocessing unit 2 with a known standard character pattern. It is equipped with

Among these, the scanner 1 includes an optical system 5 consisting of illumination, lens filters, etc. (not shown), and a photoelectric system 5 that reflects light from the illumination on a paper surface, receives reflected light that is collected by a lens through a filter, and converts it into electricity. A CCD sensor 6 as a conversion element, a preamplifier 7 that amplifies the analog signal output from the CCD sensor 6, and an analog-to-digital conversion (A/D conversion) of the output signal of the preamplifier 7 into a quantized electrical signal. and an A/D converter 8 for conversion. In this embodiment, a two-dimensional CCD sensor, which is a two-dimensional image sensor, is used as the CCD sensor 6, and the two-dimensional scanning of the character image is performed by the CCD sensor 6.

The preprocessing unit 2 includes a first binarization unit 9 that binarizes the quantized electrical signal output from the scanner 1 using a threshold value that follows the white level of the background of the paper, and a first binarization unit 9 that binarizes the quantized electric signal output from the scanner 1 using a threshold value that follows the white level of the background of the paper surface. Second binarization means 1 that binarizes using a threshold calculated based on
0, a third binarization means 11 that binarizes with a threshold value determined empirically in advance, and these three binarization means 9, 10 .
The first . Second. The third image memory 12.1314 and the image signals stored in each of these image memories (hereinafter simply referred to as "
The preprocessing means 15 performs noise removal and deformation correction of the image pattern (referred to as "image pattern").

The character recognition unit 3 stores a dictionary memory 16 in which a predetermined standard character pattern is stored in advance, and performs matching between the standard character pattern stored in the dictionary memory 16 and the image pattern of each input character from the preprocessing unit 2. and a recognition control section 11 for recognizing input characters.

Furthermore, in this embodiment, the output stage of the recognition control section 17 is provided with a comprehensive judgment section 4 that comprehensively judges the results of character recognition corresponding to each binarization.

Next, the overall operation of the above embodiment will be explained.

The image signal captured by the CCD 6 sensor inside the scanner l and subjected to photoelectric conversion is amplified by the preamplifier 7,
The signal is then A/D converted by an A/D converter 8 and converted into a quantized electrical signal. The quantized electric signal (digital value) output from this A/D converter 8 is
The signals are inputted to binarization means 9 to 11 of . These quantized electrical signals are subjected to parallel binarization processing by these binarization means 9 to 11. That is, the first binarization means 9 performs binarization using a white level tracking type threshold method, the second binarization means 10 performs binarization using a black level tracking type variable threshold method, and the third binarization means 9 performs binarization using a black level tracking type variable threshold method.
The digitization means 11 performs binarization using a fixed threshold method. Then, these binary quantized image signals are
．． Second. The image patterns are two-dimensionally stored in third image memories 12, 13, and 14, and are subjected to preprocessing such as noise removal and deformation correction by preprocessing means 15 prior to recognition processing.

After that, the recognition control unit 17 performs matching between the standard character pattern stored in the dictionary memory 16 in advance and each image pattern from the preprocessing unit 2, and recognizes the input character.
The result is output to the comprehensive determination section 4 at the next stage. Then, the comprehensive judgment section 4 finally judges the character recognition results by three types of binary conversion by majority vote.

〔Effect of the invention〕

As described above, according to the present invention, the preprocessing section includes the first binarization means that binarizes the quantized electrical signal output from the photoelectric conversion section using a threshold value that follows the white level of the background of the paper surface. , a second binarization means that binarizes with a threshold value calculated based on the blackest mesh of surrounding pixels, and a character recognition unit that integrates the character recognition results corresponding to each binarization. Since it is equipped with a comprehensive judgment unit, it is possible to process multiple binarization methods in parallel. It becomes possible to recognize faint and blurred characters without erasing them.

Therefore, both dark characters and blurred characters can be recognized more accurately, which increases the recognition rate of standard characters (80%) and eliminates the conventional problems of misreading and rejecting as unreadable characters. It is possible to provide an unprecedented and excellent optical character reading device that can reduce the number of characters.

[Brief explanation of the drawing]

FIG. 1 shows a block opening showing an embodiment of the present invention. 1...Scanner as a photoelectric conversion unit, 2...
...Preprocessing unit, 3...Character recognition unit, 4...
. . . Comprehensive judgment unit, 9 . . . First binarization means,
10...Second binarization means.

Claims (1)

[Claims] (1) A photoelectric conversion unit that photoelectrically converts reflected light obtained by optically scanning characters on a paper surface, a preprocessing unit that performs binarization of the output signal of this photoelectric conversion unit, and this preprocessing unit In the optical character reading device, the preprocessing unit includes at least a first binarization means that binarizes the quantized electrical signal output from the photoelectric conversion unit using a threshold value that tracks the white level of the background of the paper; The second step is to binarize using a threshold value.
1. An optical character reading device comprising: a binarization means, and the character recognition unit further includes a comprehensive determination unit that comprehensively determines the results of character recognition corresponding to each binarization.