JPH0135384B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention particularly relates to a pitch extraction circuit used in an optical character reading device that reads text in a free format.

Conventionally, in optical character reading devices (hereinafter referred to as OCR), which specifically read text, it has been difficult to use a format control method that specifies character pitch in advance, so it has been necessary to extract character pitch by OCR itself. By the way, in speech recognition, accurately extracting the fundamental period of speech is
extremely important. Autocorrelation methods and the like have been developed as extraction methods, and it is well known that they are quite effective. Also, extracting the fundamental period in speech recognition, and 1 in OCR.
There is a correspondence between cutting out characters one by one (hereinafter referred to as "cutting"), and both are essential techniques.

In the case of character recognition, it is possible to extract character pitch with a fairly high degree of accuracy by using the space between characters, but the algorithm for character pitch extraction becomes complicated to deal with noise, separated characters, etc. The drawback was that it was not possible to raise the level sufficiently.

This invention was made in view of the above-mentioned circumstances, and is an optical character reader that can relatively easily extract pitches such as character pitches from free-format text written on a form with a certain degree of accuracy. The object of the present invention is to provide a pitch extraction circuit used in a device.

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a schematic configuration diagram including a pitch extraction circuit according to an embodiment of the present invention. In the figure, 1 is a photoelectric conversion unit that optically scans a read form with characters etc. It has the function of converting light reflected from the surface into electrical signals. Reference numeral 2 denotes an A/D (analog/digital) converter, which has a function of converting the analog electrical signal sent from the photoelectric conversion section 1 into a multi-value digital signal. Reference numeral 3 denotes a line buffer, which has a function of storing the multivalued digital signal sent from the A/D converter 2 for one line of a form. That is, the line buffer 3 stores a multi-value quantized pattern sequence for one line of a form. 4 is a projection circuit which has a function of extracting the multi-value projection value in the vertical direction (Y direction) of the multi-value quantization pattern string for one row stored in the line buffer 3 and storing it as the projection value P(x). I have it. Reference numeral 5 denotes an autocorrelator, which has a function of calculating an autocorrelation function φ(τ) of the projection value P(x) obtained by the projection circuit 4. 6 is a peak detection circuit that detects the peak of the autocorrelation function φ (τ) obtained by the autocorrelator 5, determines the fundamental period (pitch), and checks whether the peak value is within a reasonable range for character pitch. It has a function to check etc. and output the final pitch information. This pitch information is supplied to a check section (not shown) as information for checking the multi-level quantized pattern string stored in the line buffer 3 into patterns for each character.

Next, the operation of the above embodiment will be explained. For example, a form CH on which characters are written as shown in FIG. 2 is scanned by the photoelectric conversion section 1. For example, the character string entered in the first line of the form is photoelectrically converted, multi-value quantized, and stored as a multi-value quantized character pattern string in the line buffer 3 as shown in FIG. . From the character pattern string stored as shown in FIG. 3, the projection circuit 4 extracts multilevel projection values in the vertical direction (Y direction) as shown in FIG. At this time, the value with the highest quantization level in the Y direction is set as the projected value. Then, each projection value in the X direction is stored as P(x). In addition,
If precision is not required, binary projection may be used instead of multilevel projection. Further, the scan interval should normally be set to about 0.1 mm.

An autocorrelator 5 obtains an autocorrelation function φ(τ) for the projection value P(x) obtained by the projection circuit 4. This autocorrelation function φ(τ) is the projection value P
(x) is defined by the following equation.

φ(τ) = lim T→∞1/2T∫ ^T _-T P(x)P(x+τ)dx The projected value P(x) is actually a discrete value, and the analysis interval is within 3 line buffers, so , the autocorrelation function φ(τ) for the projection value P(x) can be obtained using a simple autocorrelation calculator 5.

Then, a peak detection circuit 6 detects a peak from the autocorrelation function φ(τ) obtained by the autocorrelator 5, and checks whether the detected peak is within an appropriate range for the pitch of a character. After that, the final pitch information is output to an inspection section (not shown). When detecting peaks in this peak detection circuit 6, separation from pseudo peaks is important.
For example, when pitch is extracted from speech, if the speech waveform P(t) is as shown in FIG. 5A, its autocorrelation function φ(τ) becomes as shown in FIG. 5B. The autocorrelation function φ of this speech waveform
In (τ), in addition to the peak Pa corresponding to the fundamental period, a pseudo peak Pb corresponding to formant information appears. This is because the formant information convorooted (weighted) to the fundamental period has a large value as a pseudo peak. In the case of speech, this pseudo peak causes erroneous detection, but in the case of text, there is less possibility of erroneous detection because it does not have a frequency component corresponding to a formant. Furthermore, since the disturbance in character pitch can be estimated to some extent from the peak value, effective information can be provided to the verification section.

Note that in the case of British sentences, the spaces between words are taken quite accurately, so it is also possible to extract pitches for each word.

Furthermore, in the above embodiment, the autocorrelator 5 is used to obtain the autocorrelation function φ(τ), but a fast Fourier transform circuit is used to obtain the frequency power spectrum P(w), and as an inverse transform, the autocorrelation function φ(τ) is obtained. The function φ(τ) may also be determined. in this case,
A device that combines a speech recognition or synthesis device with OCR has the advantage that the fast Fourier transform circuit can be used in common with the speech recognition or synthesis device and OCR.

Furthermore, as shown in FIG.
etc., errors may occur due to character pitch extraction. Therefore, when calculating the projection value P(x) in the projection circuit 4,
In the above embodiment, the highest projection value in the vertical direction (Y direction) is set as the projection value P(x), but as shown in FIG. 7, the accumulated value of the projection values in the vertical direction is the cumulative projection value P (x), the portion h affected by the ghost becomes relatively small, as shown in FIG. 7, and the error can be reduced. Furthermore, as shown in Fig. 6, when taking the vertical projection of the character pattern, if the black-white-black part is made black-black-black, the part h affected by ghosts can be further reduced relatively. It can be made small and effective.

Therefore, in such a pitch extraction circuit,
There is no need to create a complicated algorithm, and by adding a relatively simple circuit, character pitches can be detected with relative accuracy, and pitch information can be provided effectively, especially in OCR, which reads text. Note that the characters entered on the form may be printed or written. It is also effective for items that are arranged in a regular manner, such as numbers, hiragana, kanji, and figures, but when it comes to items that are not arranged properly, a certain amount of information can be given to the cutting unit, leading to improved cutting accuracy.

As described above, according to the present invention, the pitch is used in an optical character reading device that can relatively easily and with a certain degree of accuracy extract pitches such as character pitches from free-format text written on a form. An extraction circuit can be provided.

[Brief explanation of drawings]

FIG. 1 is a schematic configuration diagram of an embodiment of the present invention, FIG. 2 is a diagram showing an example of filling in a form, FIG. 3 is a diagram showing a character pattern string taken into a line buffer, and FIG. Figure 3A is a diagram showing the multilevel projection of the character pattern string, Figure 5A is a diagram showing the audio waveform, Figure 5
Figure B is a diagram showing the autocorrelation function for the above speech wave, Figure 6 is a diagram showing a character pattern string affected by ghosts, and Figure 7 is a diagram showing the cumulative projection value of the character pattern string shown in Figure 6. It is. 1...Photoelectric conversion section, 2...A/D converter, 3...
... line buffer, 4 ... projection circuit, 5 ... autocorrelator, 6 ... peak detection circuit.

Claims (1)

[Claims] 1. A photoelectric conversion means that uses characters or figures written on a form as a quantization pattern, a storage unit that stores a quantization pattern for one line of the form, and a quantization pattern for one line stored in this storage unit. A projection circuit that takes a vertical projection of A pitch extraction circuit characterized by comprising a detection circuit.