JP4280939B2 - Position plane image recognition computer software - Google Patents

Description

The present invention relates to reduction in processing time and accuracy of image recognition.

There are (Non-patent document 1) and (Non-patent document 2) as literature on sensing technology.
By Sakai Yukiichi Introduction to Image Processing and Pattern Recognition Morikita Publishing Co., Ltd. 2006/12/20 First Edition Written by Hiroyuki Kitayama. NET Framework programming technique for Visual Basic / C # Vol. 5 Graphics & Image (I) C # edition Cut System Co., Ltd.

In the image identification method using Fourier and the method of identifying the pixel position of the image by converting it to a single numerical data, it is often difficult to make an identification decision. To solve this problem, a more flexible method However, if a search is performed in units of pixels, it takes time and is slow. Develop computer software that can recognize images accurately with high processing speed

1, 4, 5, 6 and 7 are very different from the actual ones. In addition, the position plane system image recognition computer software of the present invention may be called character conversion software.

Phase 1 is collated with a conventional Fourier image identification method or a method of identifying the position of the pixel of the image as a single numerical data, and as phase 2, the BMP format image is directly read into the array in bytes, In the case of the ASCII code, we have developed systematic computer software that searches for numerical values larger than those around code 126, matches the images, and identifies them.

When used in computer software that converts compressed character images, such as BMP format, into text characters that are commonly used on computers, not only will it be able to convert correctly, but it will also be a bit complicated if you make the database a little more complicated. It seems that it becomes possible to cope with the difference in fonts without registering characters. In other words, if the images of the expanded characters are sorted in order of increasing area and saved in the database, there are quite a few cases where it is not necessary to register new characters. If it is a specific handwriting, there is a possibility that handwritten characters can be verified.

I think that it is most suitable for computer software that converts text images such as BMP format into text characters that are generally used on computers.

Both Phase 1 software and the position plane system image recognition computer software of the present invention were performed without aligning the images. There is also a problem that information other than image information of files written in BMP or JPEG format is included. It is still unclear whether the white part should be compared or whether the black part should be compared with data including file information other than images. Because I confirmed it with two letters, but it wasn't understood for some reason, it was exactly the same numerical difference. In addition, it was carried out without processing dust and smudges when converting documents and scanner files. I think that it is important to add these processing functions.

The present invention also performs high-speed processing, but it takes a considerable amount of time in other parts. In addition, in order to reduce the time required for sensing, I think it is necessary to use computer software that directly accesses memory using C ++ and C # pointers.

As shown in <Explanatory drawing of equipment used in the embodiment> in FIG. 1, a computer main body indicated by reference numeral 2, a keyboard indicated by reference numeral 4, a mouse indicated by reference numeral 5, a monitor indicated by reference numeral 1, and a scanner indicated by reference numeral 6. It was implemented using a computer system consisting of a printer with symbol 3 and all the equipment connected to the computer body with symbol 2 with cables.

The software code is described below. The development compiler was coded using Microsoft Visual Studio 2005 from Microsoft Corporation. The computer of Fig. 2 used a DOS / V general-purpose machine. The scanner of Fig. 6 used the Seiko Epson Corporation inkjet printer (multifunction machine) PM-A820. The OS used was Windows XP Professional. Microsoft Word2000SR-1 was used to create the manuscript. Font characters are Cleo's brush blame Ver. We used 14 HGP Soeimaru pop bodies. The page setting specifies only the number of lines, the number of lines is 36 lines, the margins are all 35mm above, below right and left 30mm header 15mm footer 17.5mm. For the font size, we created a manuscript using the HP Deskjet 970Cxi ink jet printer of Hewlett-Packard Co., Ltd., with the reference symbol 3 as a standard setting. 11 manuscripts were prepared in A4 size with only one letter of the letter A in the alphabet printed. In the same way, we prepared 11 manuscripts of alphabet B, alphabet C and alphabet D. This manuscript was set in the scanner with symbol 6 and scanned to create a BMP format image file and transferred to the computer with symbol 2. The scanner was scanned in professional mode with a resolution of 300 dpi and the image type in monochrome. This image file is recorded on the built-in hard disk of the computer body as shown in FIG. It is used to register each ABCD image of the scanner image recorded on the internal hard disk in the database. An experiment was conducted to convert 10 ABCD images of the remaining scanner images into text characters. However, since it is a question of the accuracy of processing, it must be compared.

Therefore, we compared the computer software for Phase 1 only with the computer image recognition computer software for the position plane system of the present invention. All the manuscripts used were exactly the same as those of the position plane system image recognition computer software of the present invention. In other words, the scanner image created by scanning the document using the scanner with the symbol 6 was used as it was. The scanner image that was first registered in the database was registered with the same image. The comparison results will be described later. The program code of the position plane system image recognition computer software of the present invention is described below, and then the operation of the position plane system image recognition computer software of the present invention is explained in sequence using the figures.

Form1. Designer. cs file

Form1. cs file

Form2. Designer. cs file

Form2. cs file

Form3. Designer. cs file

Form3. cs file

Program. cs file

The above file was built to create an exe file (position plane image recognition computer software of the present invention). I installed this computer software and started it on the internal hard disk of the computer. Next, please see <Simple flow chart explanation of database registration procedure in the position plane system image recognition computer software of the present invention> in FIG. When this software is started, step 100 is executed first. If characters are registered in the database, the full path of the location where the characters, single digitized character font data, and the matching image file are stored, and an expanded image are created and the data is stored as the matching image file. Assign the serial number required for saving to the base to an array or variable. The single digitized character font data read here reads the BMP format file as it is into the array in byte units, searches the array for numerical values larger than around code 126, and subscripts of the array Is regarded as the position of the image, and the total number of the array subscripts is used.

Figures 2 and 3 show the image files for verification. This is an expanded image. In the following, an expanded image is called an expanded image or a reduced image is called a reduced image.

Then go to step 101. The operation screen of the position plane system image recognition computer software of the present invention shown in FIG. 4 is displayed. When the operation screen is displayed, it is step 102. Clicking the menu for registering characters at the top of the operation screen displays the following menu in drop-down format. The next menu is also registered. Double-click with the mouse to register this character. Step 103, the character registration operation screen of Fig. 5 is displayed. Step 104, Left click the read button of the character file to be registered with the mouse and browse the internal hard disk to specify the file. When a file is specified, an image is displayed in the image display box indicated by the reference numeral 7. Step 105, image display is automatically performed by computer software program processing, not by human operation. Step 106: The characters displayed in the image display box are visually checked, and the characters corresponding to the image are manually entered into the text box using the keyboard. Step 107: Create character font data by calculating the single digitized character font data described above. Create an image with expanded characters and save it in place. At that time, a serial number is added to the file name. Then, the data including the full path of the save destination and the serial number and the character font data are saved in the database. If you do not register at least three types of character font data, an error will occur in the character conversion software. Step 108, If you want to end the registration process, left-click the close button with the mouse. When the character registration operation screen is closed, the flow starts from step 110 in the simple flowchart explanatory diagram of the position plane system image recognition computer software of the present invention in FIG. The computer software program automatically displays the operation screen of the position plane system image recognition computer software of the present invention shown in FIG. Step 111, Left-click the scanner image file loading button to be replaced with the txt character, and specify the file by referring to the internal hard disk. Then, the image is displayed in the image display box indicated by symbol 7 by the computer software program. When displayed, left-click the conversion execution button with the mouse. Step 112 enters the process of phase 1 in FIG. This is the phase in which approximate values are found using a single digitized data of character positions. The next step 113 enters phase 2. First, find out three types of character font data in order of approximate values in Phase 1. Then, check in order. Since the file saved as the image file for verification is an image of expanded characters, first reduce the characters. The image of the contracted text and the stored expanded text is superimposed and saved as a single file. Then, calculate the total number of pixels in the white part of the expanded character image saved as the matching image file, and subtract this from the total number of pixels in the image file to obtain the number of pixels in the character part. Calculate the sum of. Similarly, find and compare the sum of the character parts of a character image created by overlaying a new character image. Then, it is judged that the character that was the approximate value is correct, and the character is displayed in the conversion candidate character text box. Next, the confirmation screen for the character conversion result shown in Fig. 6 is displayed. If the conversion candidate character is correct, left-click on the OK button. Proceed to step 118. Characters are displayed in the text box converted to txt. Next, learn the correction of approximate values and save the corrected character font data in the database. If you want to quit the character conversion software, left-click the quit button with the mouse of the symbol 5 and quit. Now, if the conversion result is not correct on the character conversion result confirmation operation screen in step 114, proceed to step 115 in this case. The list of conversion candidate characters is displayed on the operation screen of the position plane system image recognition computer software of the present invention shown in FIG. At the same time, the characters displayed in the conversion candidate character text box are deleted. Figure 7 shows this state. Step 116, find the correct conversion character from the list of conversion candidate characters, and left-click with the mouse of FIG. Then, the specified character is displayed in the conversion candidate text box. When it is displayed, left-click the conversion confirmation button with the symbol 5 mouse. Step 117: The expanded character image is saved as a file name with a serial number assigned to the character at a predetermined location. Include the full path and serial number of the file, character font data, and expanded character image, and save them in the database. Because it seems that there was no correct answer. Step 119, if you want to finish, left-click the end button with the mouse of the symbol 5 to finish.

The results were as follows for Phase 1 only computer software.
In the case of A, the number of times correctly converted 7 times The number of times mistaken in B The number of times mistaken in D 2 times in the case of B Number of times correctly converted 4 times The number of mistakes in A number 6 times C The number of times correctly converted 10 In the case of time D: Number of times of correct conversion 3 times Number of mistakes in A 1 Time of mistakes in B 6 times In the case of the position plane system image recognition computer software of the present invention, it was as follows.
In the case of A The number of times of correct conversion 10 times In the case of B The number of times of correct conversion 10 times C The number of times of correct conversion 10 times D The number of times correctly converted 9 times The number of times mistaken for B One or more results Therefore, I was convinced that the position system image recognition computer software of the present invention can be said to be a computer software that can be expected to improve considerably the accuracy.

It is explanatory drawing of the equipment used in the example. It is a simple flowchart explanatory diagram of the database registration procedure in the position plane system image recognition computer software of the present invention. It is simple flowchart explanatory drawing of the position plane system image recognition computer software of this invention. Operation screen of the position plane system image recognition computer software of the present invention. This is the character registration operation screen. This is the confirmation screen for the character conversion result. This is the operation screen of the position system image recognition computer software of the present invention that has changed since the confirmation of the character conversion result.

Explanation of symbols

1: Monitor 2: Computer main unit 3: Printer 4: Keyboard 5: Mouse 6: Scanner 7: Image display box 8: Database 9: Computer software program processing 10: Manual manual input 11: Phase 1
12: Phase 2

Claims (2)

In phase 1, conventional OCR technology, for example, an image identification method using Fourier or a method for identifying the position of a pixel of an image as a single numerical data, etc. Several candidates to be converted into text characters to be used are automatically selected, and in phase 2, previously composed of a collation image database consisting of collation image files and tesquite characters corresponding to each of the collation image files The created database is created by cutting out only the characters from a character image such as BMP format that has been read in order to convert them into text characters that are generally used on computers. Create an expanded character image file and save it as a verification image in the verification image database. The text characters corresponding to the characters written in the image file for verification are artificially entered and recorded in the database, and the text characters generally used on computers from character images such as BMP format. In order to convert the image into a scanner image, only the characters are cut out from the scanned scanner image file, a new character image file is created by reducing the character image, and the reduced character image file is stored in the collation image database. A new character image file is created by superimposing a dilated character verification image file, and the character image of the character image file newly created by superimposing these two images If the area is calculated and this is used as the area data, and an example is given, for example, a character image in which the characters are written in black and the background is written in white In this case, all the image file data is read into the array in byte units, and the array element that either black or white color code in the array is supposed to be read is searched. By calculating the number of array elements, a single numerical data that captures the character image on the surface is calculated, and the area data and the above-mentioned character image file based on the image data for verification are simulated. By calculating the area of the character by the same method as the above-mentioned calculation method, a single numerical data that captures the character image file of the image data base for collation is calculated, and both of them, that is, 2 Area data of character image of character image file newly created by superimposing image files of image and area of character image of character image file of character image file for collation Calculation is performed on each of the verification image files of the conversion candidates selected in phase 1 stored in the verification image database, and from among these, the verification image file is calculated. Automatically narrow down to a single candidate that can be judged to be an exact match or an approximate value, and manually input the text characters corresponding to the candidate and the text characters corresponding to the character image saved in the image database for verification A computer software program that converts text images such as BMP into text characters that are commonly used on computers by selecting tesquite characters from a database that has been created in advance, or this computer software program Including computer software Create a database consisting of collation image database consisting of collation image files and a database consisting of tesquite characters corresponding to each of the collation image files in advance. A character image file is created by cutting out only the characters from the BMP format character image that has been read in order to convert them into characters and expanding the cut characters, and this is used as a verification image in the verification image database. In addition to saving, text characters corresponding to the characters written in this image file for verification are manually entered and recorded in the database, and then the character image in BMP format etc. To convert to text characters to be used, only the characters are cut from the scanned scanner image file and Create a new character image file with the character image reduced, and superimpose the reduced character image file with the expanded character verification image file stored in the verification image database. If you create a character image file, calculate the area in the pseudo meaning of the character image of the newly created character image file by superimposing these two images, and use this as area data, for example For example, in the case of a character image in which characters are written in black and white in the background, all image file data is read into the array in units of bytes, and one of the black or white color codes in the array Find the array element that is supposed to be read, calculate the number of array elements, calculate the unified numerical data that captured the character image in the plane, this By calculating the product data and the area of the character in a pseudo sense of the collation image data base character image file by the same method as the above calculation method, the collation image data base character image file is displayed. Calculate the unified numerical data captured in step 2 above, that is, the area data of the character image of the character image file newly created by superimposing the two image files, and the character of the image data base for verification The calculation to obtain the difference from the area data of the character image of the image file is performed for all the matching image files stored in the matching image database, and from there, the matching to the matching image file is complete. It automatically narrows down to one candidate that can be determined to be an approximate value, and the text characters corresponding to the candidate are converted into a matching image database. By manually inputting text characters corresponding to the character image saved in the database and selecting a tesquite character from a pre-created database, it is generally used on a computer from character images in BMP format etc. Computer software program that converts to a text character or computer software that includes this computer software program

Images