JPS6037952B2 - Optimal binarization method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an optimal binarization method for converting multi-value video information into binary video information so as to maximize the figure recognition rate.

In an OCR character recognition device, characters handwritten or printed on a sheet of paper are optically scanned and converted into a multivalued gray level video signal to obtain a character pattern for recognition.

However, inside the recognition circuit, for ease of handling and simplification of the device, in most cases a black and white binary character pattern is used. In this case, line width variations such as blurring and blurring of characters may occur depending on how the threshold value (slice level) for binarization is selected. In this way, how to take the function value in the case of binarization becomes an important parameter depending on the recognition. The present invention
This invention is based on the above considerations, and the object is to provide an optimal binarization method with a simple structure that can change multi-value video information into optimal binary video information.

Therefore, the optimal binarization method of the present invention includes a video buffer in which a multilevel data signal is stored, and a slice circuit with a variable slice level that binarizes the video signal extracted from the video buffer. , a means for slicing the multilevel video information at different slice levels and converting it into a binary video signal, and a black V for each of the plurality of binary video signals created by slicing the multilevel video information at different slice levels. means for determining a line width instruction rate by dividing the number of points by the surrounding number; and a setting signal for setting the slice level of the variable slice level slice circuit based on the plurality of line width instruction rates and the reference line width instruction rate. The device is characterized in that it has a sending means for sending out the information. The present invention will be explained below with reference to the drawings. FIG. 1 shows one scanning line segment of a photoelectrically converted character video signal, and FIG. 2 shows a binarized pattern of a printed character "Island" when the slice label is changed. Figure 2 A is the character pattern when it is converted into 200 million slices at the slice level ■, and the opening in Figure 2 is the character pattern when it is binarized at the slice level ■.
FIG. 2C shows a character pattern when binarized at slice level ■. From FIG. 2, it can be seen that the character pattern becomes crushed or blurred by changing the slice level. By the way, when performing such binarization, the character pattern changes depending on the slice level, so the problem is which slice should be used for binarization, but this depends on how the recognition dictionary is created. The question is whether it was done.

The recognition circuit must be able to recognize characters that are blurred or blurred, but it is better to create a dictionary with a binary pattern as shown in Figure 2 (A). Although the recognition rate is high even for character patterns such as
It is natural that the recognition rate is high for a blurred character pattern as shown in FIG. 2A, but the recognition rate is low for a blurred character pattern as shown in FIG. 2C. Therefore, if the dictionary pattern is made with a line width like the one shown in Figure 2, the input pattern to be taken can be binarized as in the line width shown in Figure 2. Desirably, such binarization is an optimal binarization method. Therefore, in order to determine the slice level for obtaining such optimal binarized video information, the present invention uses the following value W.
is based on.

The number of black dots is the number of black meshes in the text part of the binary video, and the number of surroundings is the number of boundaries between black (text part) and white (paper surface).

Even if you change the slice level of the binarization and try to change the line width, this number is as long as it does not cause blurring, where the character part is completely lost, or collapse, where adjacent characters stick together. , the change in value is small, and compared to the change in the number of sunspots, it is 1'1 star degree. In this way, since the number around the denominator of the value W hardly changes depending on the slice level, the value W can be considered to approximately represent the line width. Hereinafter, W will be referred to as the line width instruction rate. FIG. 3 shows how the value W changes depending on the slice level.

Figure 3 shows the slice level of several lines of printed characters.
There is a plot of the value of W when binarized with ■, ■, ■ for each slice level, and it is found that there is a very clear correlation between the slice level and the line width indication rate W, and also the character It can be seen that there is not much change depending on the row or row. That is, it can be seen that the line width clearly changes depending on the slice level. Therefore, it is considered that which of the plurality of slice levels should be binarized depends on whether the dictionary pattern is created centering on a pattern with a line width such as .

FIG. 4 shows this situation. In FIG. 4, the value of the line width instruction rate W of the dictionary pattern is set to 1.5. As is clear from FIG. 4, the recognition rate changes depending on the slice level, and in this case, slice level (3) has the highest recognition rate. FIG. 5 is a block diagram of one embodiment of the present invention, in which 1 is a video buffer, 2-3 is a slice circuit, 3-0 is not present, 3-3 is a sunspot counting circuit, 4 10 is blank, 4-3 is a peripheral point counting circuit, 5-10 to 5-3 are division circuits, 6 is a comparison circuit, and 7 is a slice circuit, respectively.

The slice circuits 2-0 and 2-3 have different slice levels, and the slice level of the slice circuit 7 can be changed. The reference line width instruction rate is, for example, the number of black V points/number of surroundings of a dictionary pattern. A multi-level video signal sent from an observation section (not shown) is stored in a video buffer 1 and is also sent to slice circuits 2-0 to 2-3. The multilevel video signal sent to the slice circuit 2-0 is binarized at a predetermined level, and the number of black V points of this binarized video signal is counted by the black V point counting circuit 3-0, and the number of surrounding points is counted. are counted by the surrounding point counting circuit 4-0. The division circuit 5 divides the input number of black dots by the number of surrounding points to obtain a line width indication rate WI. Similarly, the line width instruction rates W2, W3, and W4 are obtained for each video information binarized by the slice circuit 2-1 and the slice circuit 2-3. The comparison circuit 6 compares the reference line width instruction rate W and the twill width instruction rate W1, W2, W.
3. Compare W4 and select W from W1, W2, W3, W4.
Wm closest to Wm is selected, and the slice level value corresponding to that value Wm is sent to the slice circuit 7. Slice circuit 7
The slice level is set to the slice level sent from the comparator circuit 6, and the multilevel video signal extracted from the video buffer is binarized using this slice level. In Figure 5, the video buffer may be for one character or one line, and even if the value of W is determined for each line, the W value for each line is approximately at one location, as shown in Figure 3. It can be seen that an optimal binarized video can be obtained in which the line width indication rate W is not completely the same as the line width indication rate W of the dictionary because of the clustering, but the line width indication rate W is approximately the same value. Furthermore, the slice level value corresponding to the reference line width instruction rate W can also be determined by interpolation.

[Brief explanation of the drawing]

1 is a diagram showing the waveform of one scan of a photoelectrically converted character video signal, and FIG. C is a diagram showing character patterns when binarized at different slice levels,
Figure 3 is a diagram showing how the number of black V points/number of surroundings changes depending on the slice level, Figure 4 is a diagram showing the relationship between recognition rate, slice level, and number of black points/number of surroundings, and Figure 5 is a diagram showing how the number of black V points/number of surroundings changes depending on the slice level. 1 is a block diagram of one embodiment of the present invention. FIG. 1...Video buffer, 2-0 absent, 2-3...slice circuit, 3-0 absent, 3-3...sunspot number circuit, 4-10 or 4-3...surroundings Point counting circuit, 5-0
Or 5-3... Division circuit, 7... Slice circuit. 1 figure, 2 figures, 3 legs, 1 figure, 2 figures, 3 legs, S legs

Claims (1)

[Claims] 1 a video buffer in which a multilevel video signal is stored;
a variable slice level slicing circuit for binarizing the video signal read from the video buffer; a means for slicing the multi-level video information at different slice levels and converting it into a binarized video signal; Means for determining a line width indication rate by dividing the number of black points by the number of surroundings for each of a plurality of binary video signals created by slicing at the slice level, and the plurality of line width indication rates and a reference line width indication rate. and means for transmitting a setting signal for setting the slice level of the slice circuit of the variable slice level based on the above.