JPS61156484A - Picture image processor - Google Patents

Abstract

PURPOSE:To carry out appropriate binarization processing by carrying out pre-processing for recognizing information read by reading means with the aid of multivalued digital data. CONSTITUTION:Picture images in a character frame position detection area are taken out sequentially out of a multivalued picture image data 31 which is an output data from an A/D converter 23, and transferred to adders 32 and 35. When picture taking out is of a luster scan system in an x direction, density of a y direction projection distribution is available by adding density distribution in one line, and addition completes by reciprocating scanning between the adder 32 and a latch 33, thereby storing said distribution into a memory 34. As for the x direction, scanning reciprocates between the adder 35 and a memory 36. A threshold value deciding circuit 39 decides a threshold value from a detected frame position and the picture image data 31, and send said value to a comparator 38, thereby obtaining a secondary value picture image 30 from the picture image data 31 and the comparator 38. Thus, a character area to be recognized is extracted, and the appropriate binarization processing with respect to a character image can be carried out.

Description

[Detailed Description of the Invention] [Technical Summary] The present invention relates to a pre-processing device for an image processing device, particularly an optical character recognition device, particularly a character frame and characters written inside the character frame.
The present invention relates to a character extraction device that identifies characters and symbols and extracts only characters and symbols.

[Prior art]

In conventional optical character reading devices, the character frame on the form to be read is printed in a so-called dropout color, which is a color that cannot be distinguished from the form base using the spectral sensitivity of the reading device, so that only the text is detected. The method was common. However, in this method, due to the above-mentioned restrictions on the character frame, general writing instruments such as pencils, ballpoint pens, felt-tip pens, or PPC
etc., it is impossible to create character frames and create forms in-house. Furthermore, in the case where a dropout color is not used for the character frame, a method has also been proposed in which the character frame is detected by converting the characters and the character frame into a binary image and then performing appropriate filtering. (Unexamined Japanese Patent Publication No. 59-
(No. 77577) However, since this method performs binarization using a fixed threshold value for both characters and character frames, as shown in Figure 1, when there is a large density difference between characters and character frames, or when each character If there are variations in the concentration of
There was a drawback that the digitization threshold could not be set well, and characters were likely to be cut off or blurred.

〔the purpose〕

In view of the above points, an object of the present invention is to extract a character area to be recognized and perform appropriate binarization processing on the character without the restriction of using dropout colors for character frames etc. The purpose of the present invention is to provide a capable image processing device.

Another object of the present invention is to provide an image processing apparatus that can reliably distinguish between a recognition target object, noise, and character frames in a preprocessing stage before binarization.

〔Example〕

Next, the present invention will be described in detail with reference to the drawings. FIG. 2 shows an example of a system to which the present invention can be applied. Reference numeral 1 denotes a reader, which is capable of reading a document with a CCD or the like, editing the read image data, outputting it to the printer 3, or transmitting it to an external device via a network. Note that 2 is a control unit for OCR and image processing. Further, numeral 4 is a computer having a display, and is used to issue control commands for equipment through input from the keyboard 5.

Note that the device to which the present application can be applied may also be one of the systems having a file such as a single-source disk, a microfilm, a laser beam printer, and a workstation for controlling each device.

FIG. 3 shows a block diagram of equipment to which the present invention can be applied. The original 21 to be read is illuminated with a light source, imaged on the charging conversion element 22, converted into a time-series electric signal by raster scanning, converted into a multi-value digital signal by an A-D converter 23, and converted into a multi-value digital signal by an A-D converter 24. .25 scan direction and sub-scan direction adders and correlators separately.

The character area information is sent from the correlator 25 to a binarization circuit 26, where the image is binarized and then recognized by a recognition circuit 27. The present invention relates to preprocessing immediately before performing the above-mentioned binarization.

In FIG. 4, the adder 24 and correlator 25 shown in FIG. 3 will be explained in more detail. The procedure is shown in FIG.

As shown in FIG. 6, which will be described later, from the multivalued image data 31 which is the output data from the A/D converter 23 shown in FIG.
Sequentially extract images of the character frame position detection area, 32.35
transfer to the adder. Considering the case where image retrieval is performed using a raster scan method in the The addition is completed by going back and forth, and it is sufficient to store it in 34 memories. Regarding the X direction, the data for one line scan is stored in separate memories, and each time the line changes, it is read out and the data is added, so it is necessary to go back and forth between 35.36 It becomes K to do.

The projected density distribution obtained in the above manner is subjected to correlation calculation by 37 microprocessors, and the frame position is detected.

This part will be described later.

Based on the detected frame position and 31 image data, 39 1 darkness value determination circuits provided independently from the above determine the darkness value,
It is sent to the comparator of 38, and binary image 3 is generated from 31 and 38.
0 is obtained.

The above procedure is shown in FIG. 5, and only the X direction will be explained.

First, when the frame search area is specified (step 2), an x-direction projected density distribution is created (step 3), and then, in order to detect the peak of the density distribution, in step 4, initialization is performed.
In step 5, increase t as a parameter,
In step 6, a mask calculation is performed (correlation calculation to be described later). Then, in step 7.8, the maximum value is determined and a frame is detected.

Further, FIG. 3 will be explained in detail.

As shown in FIG. 6, the original 21 to be read has a plurality of character frames 3 whose sizes and line widths are stored in the correlator 25 in advance.
Assume that a character or symbol 32 is written in 1. The multivalued digital signals from the A-D converter 23 are sequentially added by 24 adders to create projected density distributions in the X-axis direction and the y-axis direction. FIG. 7 shows an example of the projected density distribution. The density distribution is a region where at least one character frame is included and two or more are not included, for example, point @4 in Figure 7.
Let us take the area surrounded by 1. Song I in Figure 7! Reference numeral 42 indicates an X-axis projected density distribution, and 43 indicates a y-axis projected density distribution. The length 44.45 in FIG. 7 is approximately equal to the previously known size of the character frame in the X and Y directions. The correlator shown in FIG. 325 detects the character frame position by calculating the correlation between the projected density distribution and the given mask pattern.The function of this correlator (
(inside the MPU in FIG. 4) will be explained.

In this case, correlation is the integration of a certain -dimensional function f and a given mask pattern g over a finite interval,
Refers to the operation of finding the maximum value of the integral value while shifting g little by little.

To detect the frame position in the X direction, f is the X direction projected density distribution f(x), g is a mask pattern as shown in FIG. 8, for example, and if point j is taken as the origin, then the following equation is obtained. Here, if the interval ω between the two convex parts and the size h of the convex parts are set approximately equal to the size of the character frame in the X direction and the line width of the character frame, the correlation value between f and g will be as shown in Figure 7. It first becomes maximum at t=a. Therefore, by detecting this maximum value, the frame position can be determined. If the same procedure is carried out in the y direction, the points shown in FIG. 7 can be found in the same way, so that the two-dimensional frame position point A can be detected. or,
Since the length of the frame is known, point B can also be easily detected.

Next, a method for extracting a character area after extracting a character frame will be described.

The projected density distribution is traced inside the frame from the aforementioned points A and B, respectively, and the first minimum point (arrow in FIG. 9) is taken as the character extraction range. The same applies to the other two points. The character extraction range is the area surrounded by the dashed line in FIG. As explained above, since the character area is determined using the multivalued image after A-D conversion, two
By determining the digitization threshold, proper binarization can be performed.
Further, since then, the accuracy of the slam matching method has been improved, and it has become possible to provide an image processing device with extremely high accuracy.

Since there is some redundancy in ω and h in the mask pattern shown in FIG. 8, a sufficient function can be achieved if the values are within a certain range. The shape is also as shown in Figure 10(a), for example.
Several modifications are possible, as shown in (b), and the invention is not limited to the one shown in FIG. Furthermore, when obtaining the projected density distribution shown in FIG. 7, it is not necessary to extract all points within the region 41. For example, when creating the projected density distribution in the In other words, only the concentration distribution on the dashed line 81 in the figure
It is also possible to integrate as a directional projection density distribution.

As described above, according to the present invention, when extracting images such as characters written in a predetermined frame, there is no need to use tricks such as drop-out colors, and, for example, a form made in-house using PPC can be used. is also possible. In addition, since processing is performed as a multivalued image, characters can be extracted stably by reducing malfunctions such as smearing and blurring of forms that tend to occur when an inappropriate threshold is set in binary image processing. , subsequent processing such as binarization and recognition becomes easier.

〔effect〕

As detailed above, according to the present invention, it is possible to extract the character area to be recognized without using dropout colors or the like in the character frame, and to perform appropriate binarization processing on the character image. It has become possible.

According to the present invention, since preprocessing is performed on the multivalued image before the image data is binarized, it is possible to easily perform proper binarization.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an example of a document in which there is a large difference between the character frame density and character density, or in which there is variation in the density between characters, and FIG. 2 is an example of an apparatus to which the present invention can be applied. 3 is a configuration block diagram of the device to which the present invention is applied, FIG. 4 is an explanatory diagram of the adder and correlator, FIG. 5 is a 70-chart for character frame identification, and FIG. 6 is a form diagram. Figure 7 is a diagram explaining the projected density distribution, Figure 8 is a diagram showing the mask pattern used for correlation processing, Figure 9 is the cutout range of the character area. FIG. 10 is a diagram illustrating another example of a mask pattern used for correlation processing. FIG. 11 is an explanatory diagram when scanning lines are taken every few lines in the y direction.

Claims (1)

[Claims] A reading means capable of reading frame information and image information;
An image processing apparatus characterized in that preprocessing for recognizing information read by a reading means is performed on multivalued digital data.