JPS5872276A - Optical type character reading system - Google Patents

Abstract

PURPOSE:To remove noises by converting an analog output signal from a two- dimensional sensor into digital picture information, storing the digital picture information in a picture information memory and detecting digital picture information corresponding to blanks other than characters. CONSTITUTION:A two-dimensional sensor 2 in a scanner part 1 photoelectrically converts characters projected on a sheet into analog information. The information is converted into digital picture information by a binary-coding circuit 4 in a discrimination processing part 3 and the output of the circuit 4 is stored in a picture information memory 5 and also sent to a blank detecting circuit 6, which detects blanks other than characters and noises. The output of the circuit 6 is sent to a removal circuit 7, which removes noises from the information outputted from the memory 5.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention uses a method in which characters written on paper are scanned and optically read, and the read signals are used to logically identify the characters. A method for removing analog image noise present in the picture is shown below.

An optical character reading device optically projects characters written on paper onto a group of photoelectric conversion elements, processes electrical signals from the photoelectric conversion elements, and makes decisions according to the order in which the signals occur using a dendritic logic. This is what we do. For this reason, all characters on the paper are converted directly into electrical signals for processing and judgment, but if there is any analog image noise in the form of spots or lines around regular characters, it will be difficult to read. There is a possibility that the characters may be misread or unreadable. Until now, no circuit has been proposed to remove this analog image noise.

Therefore, in view of the above-mentioned drawbacks, the present invention aims to provide an optical character reading method that removes analog image noise other than normal NAo read characters.

In order to achieve this object, the present invention uses a two-dimensional sensor to read the characters written on paper. In the character reading method for identifying the above-mentioned characters, the image information output from the upper two-dimensional sensor is digitalized. converting the digital image information into image information, storing this digital image information in the image information memory, and detecting the digital image information corresponding to the non-text margin; It is characterized by removing digital image noise contained in the information memory.

ζζ A detailed explanation of the invention will be given with reference to the figures. 1st
In the figure, a two-dimensional sensor 2 is provided within a scanner unit 5 that scans characters written on a sheet of paper. This two-dimensional sensor 2 charges and converts the projection of characters on paper into analog image information. The analog image information from the two-dimensional image sensor 2 is sent to the identification processing unit 3 to the Shinimen conversion circuit 4.
The image is converted into a digital image by . Binarization circuit 4
The output is temporarily stored in the image information buffer 5, and is processed in the margin detection circuit 6 for detecting margins other than characters and noise on the paper. Next, the output of the margin detection circuit 6 is input to the removal circuit 7, and the removal circuit 7 inputs the information from the image information buffer 5 to the margin detection circuit 6.
It processes the output from and removes noise.

The above is an outline of noise removal. The details of the noise removal method will be described below in FIG. FIG. 2 shows the state of one field of view during scanning by the scanner section 10 in FIG. This irises 2all are shown assuming 13 columns x 36 rows of pixels. Analog image information is created corresponding to this number of pixels, and then digital image information is obtained. This information can be divided into white information or black information, but for example, it is necessary to identify whether black information that becomes text information actually becomes regular text information, or whether it is noise after conversion. . A white pixel is determined without hesitation in this identification.

For example, in the analog image information shown in FIG. 3(&), as shown in FIG. 3(b), (n-1) rows, n rows, (net)
Extract rows. For example, if we focus on any one pixel, in order to determine this pixel as white, this pixel itself must first be a white pixel, and the Alternatively, at least one of the six pixels on the right must be recognized as a white pixel. In order to recognize at least one of the six pixels to the left or right of the pixel of interest as a white pixel in the immediate vicinity, if there is a black pixel to the left or right of the pixel of interest in the same row, that black pixel is identified as a white pixel. It is determined that the pixel directly to the left or right of the black pixel including the black pixel is not considered to be a white pixel. - This operation is to check whether the pixel of interest has continuity with the immediately preceding white pixel.

In Fig. 3(b), if the pixel in row n and row 1 is the pixel of interest, the pixel in row n and column 1 is a white pixel, and the pixel in row n and column 1 is a white pixel. Since the first, second, and third columns are all white pixels, this white pixel of interest is recognized as being continuous with the white pixel in the previous row. FIG. 4 shows a case where the first column of jlli and the last column are the pixels of interest. Note the diagonal l in Figure 4.
If the pixel of interest is a white pixel (marked with a circle), the pixel in Figure 4 (if the first or last column of the n-1 row is a white pixel, as shown in Figure 4 @ (b)) When the second column of n rows or the last -1 column are white pixels, and the white pixels are in the second column of n-1 rows or the last -1 column, the third column of n rows or the last -2 column as shown in Figure 4(e). When the 3rd column of the n-1 row or the last -2nd column is a white pixel, the irises 411 (a), (・), θ) are also continuous with the previous row (n-1 row) of the pixel of interest. The white pixels have a distinctive character.

Returning to Figure 2, when the 10th row, jIB column is the pixel of interest X, the 6 columns on the left side of the same row as this pixel In addition to being a black pixel, the two columns on the right in the same row as the picture 1AX are white 1ji.
As a II element, the 3rd and 4th columns are black pixels, and the 9th row in the 2nd column to the right becomes a black 1ikI element, so as a result, the target pixel X is all related to black pixels in the 9th row. Therefore, it is not possible to provide continuity as a white pixel. The same applies when pixel Yf in the 10th row and 9th column is set as the pixel of interest.

As a result of determining the presence or absence of continuity of white pixels for the pixels in 13 columns and 36 rows in the $2 diagram, only pixels X and Y are determined to have no continuity. Therefore, the white pixel group consists of pixels excluding the black pixel and the m-element XY.

In Fig. 2, the presence or absence of continuity with row n-1 was determined for the pixel of interest in row n, but then, as shown in Fig. 5, in a complete opposite of the method shown in Fig. n+ for the pixel of interest
Determine whether there is continuity with one line. As a result, 1 in Figure 5
The pixels shown in ~h are not continuous, and a continuous white pixel group is found excluding the pixels a - h and the black pixels. In other words, a reliable white pixel group is determined, and then analog image noise is determined and removed.

For this determination and removal, the selection method shown in Figure 6 is executed. A midpoint is found within each row from the continuous white pixel group shown in FIGS. 2 and 5.

That is, as shown in FIG. 6, the midpoint of white pixels indicated by O marks in the same row is determined. If there is one or two midpoints per row, then that row has no black pixels or only one black pixel. However, when there are three or more midpoints, it is assumed that there is noise other than black elements representing characters, and two of the three or more midpoints are selected. For example, looking at rows n-1 and n in Figure 6, row n-1 has two midpoints, so there is no problem, but row n has three midpoints, so reduce it by one. In this case, look at the relationship between the midpoint of row n-1 and the midpoint of row n.The midpoint on the right side of row n-1 is related to the midpoint on the right side of row n, and the midpoint of row n-1 is related to the midpoint of row n-1. The question is whether the left midpoint should be related to the center of n rows or the left and negative. In this case, select the center midpoint close to the center column. In this way, there are three or more midpoints in each row. Select two items at a time. Black pixels between the unselected midpoint and the selected midpoint are removed as analog image noise.

1 for the continuous white pixel group in Figures 2 and 5.
The midpoint is determined as shown in FIG. 617, and the midpoint is selected as shown in FIG. As a result, the characters shown in FIG. 9 can be obtained by removing analog noise.

In the explanation following FIG. 2, noise removal was performed for the number 3, but the same is true for other numbers, and two p-midpoints are always selected.

For example, when determining continuous white pixels for the number 8, the part surrounded by the black pixels of the number 8 is not considered to be continuous white pixels, so the white pixels on both sides of the number 8 are used as the midpoint. becomes 2 pieces.

As shown in Iris #2#A and FIG.
Targeting row 1 (forward) and row n+1 (return) is effective in removing analog image noise that exists only near the first column or the last column.

The above description is the processing function of the margin detection circuit 6 and removal circuit 7 shown in FIG. FIG. 10 is a more detailed diagram of these circuits. Digital image information including noise is inputted to the image information memory 5 and the margin detection circuit 6 from the two-dimensional sensor 2 shown in FIG. In the image information memory 5, digital image information including this noise is recorded in two memories 5m and 5b. On the other hand, the margin detection circuit 6 first selects the continuous white pixels shown in FIG. 2 by the top-to-bottom margin detection circuit 6a. This selection result is stored in the image information memory 6b. Furthermore, the continuous white pixels shown in FIG. 5 are selected by the bottom to top margin detection circuit 6C, and the result is stored in the image information memory 6d. 8 is an image information memory control circuit.

The removal circuit 7 selects two of the midpoints of the continuous white pixels determined by the margin detection circuit 6, and erases pixels outside the area between the two selected midpoints.

As described above in the embodiments, according to the present invention, it is possible to remove noise, extract only properly readable characters, and avoid misreading and unreadable characters.

[Brief explanation of drawings]

Figures 1 through 10 illustrate the implementation of the optical character reading system according to the present invention, where Figure 1 shows a block diagram, Figures 2 and 5 are illustrations of white pixel selection, and Figure 3-) (b) is an explanatory diagram of white pixel selection, Figures 4-) to α) are explanatory diagrams of white pixel selection of 1 row with the first and last columns as the target pixels, and Diagram 6 is an explanatory diagram of white pixel selection for midpoint selection. Explanatory diagrams, Figures 7 and 8 are explanatory diagrams for midpoint selection and midpoint cutting failure, Figure 9 is a diagram showing characters/4 turns after noise removal, and Figure 1O is a diagram similar to Figure 1. This is a more detailed explanation. In the drawing, 6 is a margin detection circuit, 5 is an image information memory, and 7 is a removal circuit. Patent Applicant Sumitomo Electric Industries Co., Ltd. Agent Attorney: Shibu Mitsuishi (and 1 other person) Figure 2 Section 3 (b) (a) (1) Published by Ci1 (2) -1--- No. 4 Figure (f) Kuchichi Shiyo Ido ■i Figure 5 Figure 6 Figure 9

Claims (1)

[Claims] Reading the characters written on paper with a two-dimensional sensor] In the character reading method for identifying the characters, the analog seal image information output from the two-dimensional sensor is converted into digital image information, and this digital image information is is stored in the image information memory, and the digital image information that appears in the margins other than the characters is detected, and digital image noise contained in the image information is removed based on the digital image information that appears in the margins. An optical character reading method that is characterized by: