JPH0432970A - Method and device for image recognition and correction - Google Patents

Abstract

PURPOSE:To discriminate a place which needs to be corrected by an operator at a glance by displaying the part to be corrected such as a place which is low in recognition rate and a place where a read image is displayed as it is by making its color different from other places. CONSTITUTION:An original drawing 4 is read by an image scanner 312, 'symbol', 'line', and 'character' are segmented from the read image and recognized, and at the same time, the likelihood ratios are found in the form of numerals. Then places which have the lowest likelihood ratio in the recognized and displayed image are displayed in a different color, one by one. An operator matches the places in the different color with the original drawing 4 and it is decided whether or not the recognition performed by a computer 1 is wrong; when it is decided that the recognition is wrong, that is corrected. This correction is performed through a keyboard 33 or inputting a correction command. Consequently, the places to be corrected can easily be searched for.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to an image recognition/correction method and an apparatus thereof, and in particular, to an image recognition/correction method suitable for easily finding and correcting areas with a low recognition rate in an image. This invention relates to a correction method and apparatus.

[Prior art] When designing printed circuit boards or semiconductor integrated circuits such as LSIs using a computer, it is necessary to draw electrical circuits, logic diagrams, etc. on drawings and input them into the computer. .

Conventionally, when performing this input, an operator of a computer views a manuscript drawing created by a logic designer or a circuit designer and inputs the input from an input device such as a keyboard. However, −
Inputting information from a keyboard or the like has the disadvantage that it takes time to input information about large-scale circuits. Therefore, it has become common to read the above-mentioned original drawings using an image reading means such as an image scanner, and to recognize the drawings using a computer and convert them into design data.

When a computer performs the above recognition, it converts the image in the drawing into binary values of white and black (generally, white is “1”).
”, black is set to “0”), a dictionary function that stores a plurality of standard images, and an image cutting function that cuts out image elements from the read image to be recognized by pattern matching or the like. A function to compare the extracted image and the dictionary, determine the degree of matching, and select the most similar dictionary image data, a function to store the name and position of the selected dictionary image data as a recognized image, and a function to store the name and position of the selected dictionary image data as a recognized image. It has a function to generate and display the entire image by combining the images.

When reading, recognizing and displaying a manuscript drawing using the above-mentioned computer, it is sufficient if the entire image can be recognized with high efficiency, but if the original manuscript drawing is scratched,
Part of the manuscript drawing is incorrectly recognized. Therefore,
Compare the image displayed on the display device with the manuscript drawing,
It is necessary for the computer operator to find the location that has been misrecognized and displayed as an incorrect image, and to specify and correct the location using a mouse or the like. At this time, if the location of misrecognition cannot be found efficiently, it will still take time.

Therefore, instead of comparing the low recognition rate with dictionary data to obtain a standard image and displaying it (which would result in an image different from the original image being displayed), the parts with low recognition rate are left as the read image. For example, in Japanese Patent Application Laid-Open No. 62-164171, when a read character image is recognized and displayed, for unrecognized characters,
The read image, for example, a scratched character image, is displayed as is.

[Problems to be Solved by the Invention] When an operator of a computer wants to modify an image displayed on a display device so that it matches the original drawing, he must search for the location to be modified. When an erroneously recognized image is replaced with an incorrect standard image and displayed, whether or not the image is an incorrect image can only be determined by comparing it with the original manuscript drawing.

Therefore, it is troublesome to search for correction points, and there is also the disadvantage that the larger the circuit, etc., the more time it takes. In this respect, as in the prior art described above, by using a scanned image with a low recognition rate as a scratched image, there is less need to compare the image with the original drawing, so it is still easy to search for a correction area. However, this search also becomes difficult as the circuit scale increases, and furthermore, if the cause of the low recognition rate is not rubbing but another cause, the search becomes even more difficult.

Furthermore, in the method in which the operator specifies the correction area using a pointing device such as a mouse, the specification process is cumbersome;
In addition, the operator must identify all the parts to be corrected by himself, and there is a risk that corrections may be omitted.

A first object of the present invention is to provide an image recognition/correction method and an apparatus for the same, which makes it easy to search for a correction location.

A second object of the present invention is to provide an image recognition/correction method and an apparatus for the same without omitting any corrections.

[Means for solving the problem] The first purpose is to read the image shown in the JJW draft drawing, recognize the image, display the recognized image on a display device, and eliminate errors in the displayed image. In an image recognition/correction method that corrects a recognized part or a part that cannot be recognized and displays the read image as it is based on input instructions from an operator, the part with a low recognition rate in the image displayed on the display device or the read image is displayed as is. This is achieved by distinguishing and displaying areas with high recognition rates from areas with high recognition rates.

The second objective is achieved by displaying a location to be modified and, when the operator instructs to modify the location, automatically displaying the next location to be modified.

[By displaying areas that the cooperator must correct, such as areas with a low recognition rate or areas where the read image is displayed as is, by displaying them in a different color from other areas, the screen can be improved. You can identify the location at a glance when you see it.

When displaying, the device side automatically searches for the correction target part based on the recognition rate, etc., and automatically displays it each time it is corrected sequentially, thereby saving the trouble of specifying the correction part and eliminating the possibility of omission of correction.

[Example] Hereinafter, one embodiment of the present invention will be described with reference to the drawings.

FIG. 2 shows an image u! according to an embodiment of the present invention. ! - It is a block configuration diagram of a correction device. This image recognition/correction device consists of a computer main body 1, an image scanner 31, and a CRT 32 that is a display device.The image scanner 31 reads an image drawn on a manuscript drawing 4, and the computer main body 1 As will be described later, this read image is recognized and displayed, corrected by the operator using the keyboard 33 and mouse 34, and then displayed on the CRT 32. Further, the corrected image data is stored in a floppy disk, which is an external storage device, as automatic design data.

The computer main body 1 includes a CPU 10, a main memory 2, an image memory 13 that stores images read by an image scanner 31, a keyboard 33. Mouse 34. l10, which interfaces with the floppy disk driver 35
12, an image memory 14 that stores image data to be displayed on the CRT 32, and a bus 11 that interconnects these.

FIG. 3 shows an example of a hand-drawn logic circuit diagram drawn on a manuscript drawing. This logic circuit diagram shows "symbols" 41 (in the illustrated example, the symbol of a Noah gate) which are circuit elements that constitute a logic circuit. , "r line" 42 connecting "symbols" and "symbols"
Character string 43 mainly consisting of alphanumeric characters indicating the "name" of "Ya r line"
Consists of etc. This logic circuit diagram is read with an image scanner, "symbols" and "names" are extracted from the read image, and the extracted image is recognized by pattern matching with a standard image, which is dictionary data, and by determining its feature amount. For a "line", which "symbol" is its starting point and ending point?
Recognize where in the body it is connected.

FIG. 4 is a block diagram of the main memory 2 shown in FIG. 2. This main memory 2 includes a program storage area (MP) 21 for storing programs that perform the above-mentioned recognition, an area (MS) 22 for storing dictionary data, and a work area (MW) 2.
3, and an area (MD) 24 for storing automatic design data.

The CPU 10 cuts out "symbols" and "names" from the image read by the image scanner 31 and stored in the image memory 13, and determines them based on dictionary data. There are various conventional methods for this extraction, but since a "symbol" is a closed figure within a certain size range, it is scanned sequentially from the upper left point of the read image, and if there is a black point (0), If you trace it and it is a closed figure that returns to its original point within a certain length, there is a high probability that it is a "symbol," so cut it out as a "symbol."

The cut out closed figure is compared with a standard image of a "symbol". There are various methods of comparison, but for example, the area of a closed figure9
Features such as the ratio of the area of the circumference 2 to the circumference, the area ratio of the 9f encircling flame rectangle to the closed figure, and the aspect ratio of the circumscribed rectangle are compared.

Alternatively, divide the cropped image into a fixed number of vertical and horizontal pixels,
Evaluate the number of pixels with matching white and black in each image (pattern matching). In either case, the degree of agreement is quantified, and when that value exceeds a certain level, it is recognized as the "r symbol".

For a "line", the "r name" of the "symbol" to which the "r line" is connected and its connection position are recognized.

The main reason for misrecognizing the ``R line'' is when the ``R line'' is cut in the middle and it is not recognized as a single line. At this time, accuracy is evaluated based on how much length is cut.

The method of recognizing a "name", that is, the method of recognizing characters, is the same as that of a "symbol". Therefore, the probability of matching for characters is also evaluated as a numerical value.

And the character string is the "symbol" near it.

It is recognized as belonging to "r line". These recognition processes are performed using the work area 23 shown in FIG.

The "symbol", "r line", and the "character string" attached thereto obtained by recognition as described above are automatic design data necessary for circuit design, etc. This automatic design data is stored in the area 24 (FIG. 4) of the main memory 2, and the CPU 10 sequentially reads out this data to
, r character string, etc. are generated and written into the image memory 14 for CRT display. A "character string" is displayed near the "symbol" or "r line" to which it belongs according to a predetermined rule. This display method is known (for example, Japanese Patent Laid-Open No. 5
No. 9-133658), and since it is not the main point of the present invention, its explanation will be omitted. In this way, the image data written in the image memory 14 is displayed on the CRT 32.

The operator compares the display image displayed on the CRT 32 with the contents of the manuscript drawing 4, and if he finds any misrecognition, he displays the "symbol",
Select "Line Jt'character" with the mouse 34 and use the keyboard 3
3, the correct data is input, the image data is corrected, and the automatic design data in the area 24 is corrected. When the modification is completed, the data stored in this area 24 is written to a floppy disk (not shown) inserted into the floppy disk driver 35. The contents of this floppy disk are read by another automatic design electronic computer (not shown), and automatic design is executed. Note that it is also possible to transfer necessary data from the image recognition/correction device to the automatic design computer using a LAN without intervening a floppy disk. Furthermore, in the case of a configuration in which the image recognition/correction device itself automatically designs, the data stored in the area 24 of the main memory 2 is used as is.

As mentioned above, all recognition results can be evaluated by numerical accuracy.6 In this embodiment, accuracy is evaluated in several stages, and the color of the image is displayed for each stage to reduce uncertainty. Make it possible to identify important parts at a glance. The procedure at this time will be explained with reference to the flowchart of FIG.

First, in step 100, a manuscript drawing is read with an image scanner. Then, from this read image, "symbol", "r-ray"
, r characters" are cut out, and the accuracy is numerically determined while recognition is performed as will be described in detail later (FIG. 5) (step 200). In the next step 101, the entire recognized result is displayed on the CRT 32.

Next, the correction process begins, and in this embodiment, in step 102 of the correction process, the colors of the parts with the lowest accuracy are changed one by one from the recognized and displayed image (symbol, line, 9 character strings). indicate. The operator compares the areas with different display colors with the manuscript drawing (step 103), determines whether the recognition made by the computer is incorrect or not (step 104), and if it is determined that the recognition is incorrect, the operator performs step 103). Corrected at 105. This modification is performed by inputting a modification command from the keyboard, and the process returns to step 102. As a result, the display color of the next location with low accuracy will be changed, and that location will be corrected.

If it is determined in step 104 that there is no misrecognition, then it is determined whether the recognition accuracy is high (step 1
07). If the recognition accuracy is high, this process ends.

If the recognition accuracy is low, return to step 102 and repeat the process.

In this way, in this embodiment, instead of the operator specifying correction points using a mouse or the like, the computer side sequentially displays points with low accuracy, so the operator can check whether the displayed points are correct or not. This allows you to concentrate on making decisions, and there is no time or hassle for making specifications.

As described above, it is possible to increase the recognition accuracy of all parts constituting the image, that is, to increase the reliability of the image.

FIG. 5 is a flowchart showing the detailed procedure of processing step 200 in FIG. First, step 201
, start scanning from the upper left point of the scanned image, and select the "symbol"
In step 202, an image that is considered to be a symbol is cut out, and after a rough recognition of the cut image is performed, it is determined whether the image is a "symbol" or not (step 203). If the image is not recognized as a "symbol" in this determination, the process returns to step 201 and the next "symbol" image is cut out. If it is recognized as a "symbol" in step 203, the "F symbol" is recognized in step 204 and the "symbol" is recognized.
It is determined what symbol is, and its recognition accuracy is calculated in step 205.

In the next step 206, it is determined whether or not there is a "line" connected to the "symbol", and if there is, the name of the "symbol" to which the "F line" is connected is determined (step 20
7) , further calculate the accuracy of the r-line and proceed to step 20.
Return to 6. The processing of steps 206 to 20g is performed on all the "lines" belonging to the "symbol", and the unprocessed r-line is
If there is no more, step 206 to step 2
Proceeding to step 09, it is determined whether or not a "character string" belonging to these "symbols" and "r line" exists. Furthermore, that "character string"
Whether or not belongs to a "symbol" or "line" depends on the "symbol",
It can be identified by whether it is within a certain distance above the "r line".

After cutting out "r character string", the character is recognized in the next step 210, the recognition accuracy is calculated in step 211, and the process returns to step 209. The above processing steps 209~ for all the character strings found in steps 201~208
After repeating step 211, step 209 to step 2
Proceed to step 12 and learn about the "symbols" and "lines" that we recognized this time.

The image corresponding to the "character string" is deleted from the image data stored in the image memory I3.

In the next step 213, it is determined whether or not black dots remain in the image memory 13, and if so, the process returns to step 201 because there is still an image to be cut out. If no black dots remain, the recognition process has been completed for all of the read images, so step 1 in Figure 1 is completed.
Proceed to 01.

Next, the process of determining the accuracy of symbol recognition (step 205 in FIG. 5) will be described. First, an example of recognizing the symbol J41 in FIG. 3 using feature amounts will be described with reference to FIG. 6. "
The first black point found by scanning in the symbol "41" is point 411. If you trace this point 411, you will come across several branches along the way, but if you trace only the inside, you will return to point 411 as shown by the arrow. Since its length is within the required length, a rectangle 412 indicated by a dotted line circumscribing this "symbol" image is first determined to be a "symbol" image. In the example shown in FIG. 6, the evaluation function can be the ratio between the area of a closed figure of the image and the area of the circumscribed rectangle, and the aspect ratio of the circumscribed rectangle. An example of the relationship between the range of feature values and accuracy is shown in the table in Figure 7.
Image recognition and its accuracy are determined by determining the accuracy for all standard images and storing the image name and accuracy with the highest accuracy.

Fifth! ! In the rough recognition processing in steps 202 and 203 in I, the accuracy of each of the plurality of evaluation functions is determined for the cut out image, and when all the accuracies are less than the value 1, it is determined that the cut out image is not a "symbol". do. It goes without saying that a rough recognition judgment can be similarly made for "lines" and "character strings," and that this judgment processing may be added to the processing procedure shown in FIG.

In the above-mentioned embodiment, each location is displayed in a different color in descending order of accuracy, but for experienced operators, all locations belonging to that accuracy may be displayed in different colors at once in descending order of accuracy. , after checking and correcting its accuracy,
All the locations belonging to the following accuracy may be displayed in different colors at once, or all the locations below a certain accuracy may be displayed at once.

[Effects of the Invention] According to the present invention, since the time required to search for a correction location is shortened, correction processing can be performed at high speed, and data reliability is improved. Furthermore, the time and effort required to specify the correction points is saved, and no corrections are omitted, further improving the reliability of the data.

[Brief explanation of drawings]

FIG. 1 is a flowchart showing the overall processing procedure of an image recognition/correction method according to an embodiment of the present invention, and FIG. 2 is a block diagram of an image recognition/correction apparatus according to an embodiment of the present invention.
Figure 3 is a handwritten logic circuit diagram that is an example of a manuscript drawing, Figure 4 is a diagram showing the configuration of the main memory shown in Figure 2, and Figure 5 is a diagram showing the recognition and recognition accuracy calculation processing steps shown in Figure 1. FIG. 6 is a flowchart showing the detailed procedure, FIG. 6 is an explanatory diagram of symbol recognition processing, and FIG. 7 is a table explanatory diagram showing an example of determining accuracy. 1... Computer main body, 2... Main memory, 4...
Manuscript drawing, 10... CPU, 13.14... Image memory, 21... Program storage area, 22... Dictionary, 23-1° work area, 24... Automatic design data storage area, 31 ...Image scanner, 32...C
RT, 33...keyboard, 34...mouse, 41
... "Symbol", 42 ... "Line", 43 ... "Name". Agent Patent Attorney Tadashi Akimoto Figure 1 Figure 2 Figure 4] Figure 4

Claims (1)

[Claims] 1. Read an image displayed on a manuscript drawing, recognize the image, display the recognized image on a display device, and identify erroneously recognized parts in the displayed image or unrecognized parts of the read image. In an image recognition/correction method in which a portion displayed as is is corrected based on an input instruction from an operator, a portion with a low recognition rate in the image displayed on a display device or a portion where the read image is displayed as is is determined as a portion with a high recognition rate. An image recognition/correction method characterized by displaying differentiated images. 2. Read the image displayed on the manuscript drawing, recognize the image, display the recognized image on a display device, and correct the correction target parts with low recognition rate in the displayed image according to input instructions from the operator. An image recognition/correction method characterized by displaying a portion to be corrected and automatically displaying the next correction portion when an operator instructs to correct the portion. 3. An image reading means for reading the image displayed on the manuscript drawing, an image recognition means for recognizing the read image, a display device for displaying the recognized image or the read image in the case that it cannot be recognized, and a display device for displaying the image on the display device. and a correction means for correcting a corresponding part of the image displayed on the display device according to an instruction input by an operator when the image displayed on the original drawing differs from the image displayed on the manuscript drawing, wherein the display device Image recognition/correction characterized by comprising: a differentiation display means for displaying areas with a low recognition rate by the image recognition means or read image display areas in an image displayed on the image, distinguishing them from areas with a high recognition rate. Device. 4. An image reading means for reading the image displayed on the manuscript drawing, an image recognition means for recognizing the read image, a display device for displaying the recognized image or the read image in the case that it cannot be recognized, and a display device for displaying the image on the display device. In an image recognition/correction apparatus, the recognition rate is improved. An image recognition system characterized by comprising means for sequentially displaying low correction target parts each time an operator gives a correction instruction.
Correction device.