JPS62157981A - Character reader - Google Patents

Abstract

PURPOSE:To miniaturize a device at low cost without damaging the accuracy of recognizing characters on a mobile object to be read by making character image read out by scanning on the same scanning line twice continuously in either odd number field memory or even number field memory the object of recognition in character recognition of a moving character image. CONSTITUTION:In the case where an input character image Di is a still picture, an MODE signal is set to '0', and accordingly, image data D1 stored in an odd number field memory 7 and image data D stored in an even number field memory 8 are read out alternately, and composed output character image D0 is supplied to recognizing process. On the other hand, in the case where input character image Di is a mobile character image, an output character image D0' obtained by duplex scanning of image data D1 of odd number field indicates the feature more sharply than the output character image D0 obtained by reading images of each field alternately. Accordingly, the MODE signal is set to '1' and the duplex read image of the odd number field memory 7 is supplied to recognizing process.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a character reading device used in the industrial field.

[Technical background of the invention and its problems]

Character reading devices used to input data into computers have developed recognition technology that extends to the area of kanji, including handwritten kanji, and have made a significant contribution to streamlining and increasing efficiency of office work.

On the other hand, even in this era of high-mix, low-volume production, the management of serial numbers of products during various production processes at industrial sites is still done using methods that are prone to input errors, such as manually transcribing them onto slips by visual inspection. This is the actual situation. Conventional general-purpose readers (OCR) read characters on managed forms, whereas in industrial settings, characters exist in complex backgrounds containing various parts. Conventional OCR has many problems such as , character position changes, etc.
This is because character reading technology cannot be applied as is.

Considering this point, character input devices in the industrial field do not use conventional OCR at the input stage of character images.
Rather than using a line sensor to store one line of characters in a single line memory and then recognize the characters, even if the characters are
Even in the case of lines, the entire screen must be stored in a frame memory using an area sensor such as a TV camera, and then character areas must be detected from the stored screen to recognize the characters.

By the way, objects to be read at production sites are in any one of a stationary state, a constant moving state, and a random moving state during various production processes. Therefore, when image input is performed using a general TV camera with an interlaced scanning method as an area sensor, there is no problem when the object to be read is stationary, but when it is moving constantly or randomly. , the input image to the frame image memory deviates between odd and even fields, and if characters are recognized as they are, recognition performance will deteriorate.

Therefore, it is possible to correct the image by calculating the amount of movement between the fields of the character image stored in the frame memory, but the image can be corrected faithfully based on the relationship between the unevenness of the image itself and the relative position of the image and the TV camera. However, it is very difficult to do so, and the disadvantage is that the circuit scale becomes large.

Additionally, an electronic shutter camera may be used to store a moving object as a still image in a frame memory, but electronic shutter cameras are more expensive than general TV cameras, and the cost of the entire industrial OCR system is low. There is one problem that cannot be achieved.

[Purpose of the invention]

The present invention has been made based on the above-mentioned problem, and even when an interlaced scanning imaging device is used as an area sensor, the recognition accuracy of characters on a moving object is not impaired, and the size and size of the image pickup device can be reduced. The purpose is to provide a character reading device that can be priced reasonably.

[Summary of the invention]

The present invention stores character images of an object to be read in consecutive odd fields inputted from an interlace scanning imaging device in an odd field memory, and stores character images of an object to be read in an even field in an even field memory, In character recognition of static character images, character images read out alternately from odd field memory and even field memory are to be recognized; in character recognition of moving character images, character images are read out alternately from odd field memory and even field memory. The present invention is characterized in that a character image read out by scanning the same scanning line twice in succession is targeted for recognition.

〔Effect of the invention〕

According to the present invention, even when image recognition of moving characters is performed using an imaging device using an interlace scanning method, an image having almost the same quality as a still image can be recognized by double scanning of an odd field memory or an even field memory. Therefore, there is no need for a special circuit for correcting the movement amount, and there is no need to use expensive means such as an electronic shutter. Therefore, it is possible to provide a character reading device that is small, inexpensive, and has high recognition accuracy.

[Embodiments of the invention]

Hereinafter, details of the present invention will be explained based on illustrated embodiments.

FIG. 1 is a diagram showing the overall configuration of this embodiment, and numeral 1 in the figure represents a product moving on a production line.

On this product 1, a character line 2 such as a serial number to be recognized is printed. This product 1 is imaged by a TV camera 3 consisting of an area sensor. As shown in FIG. 2, the TV camera 3 is of an interlaced type that performs interlaced scanning of even and odd fields. The character image from the TV camera 3 is sent to the binarization circuit 4.
It is binarized with . The digital signal from the binarization circuit 4 is sampled according to a sampling pulse output from the input control circuit 5. At the same time, address control circuit 6
is also counted up by the sampling pulse, and addresses of the odd field memory 7 and the even field memory 8 are designated alternately.

As a result, among the binarized character images, to Fig. 2 a1 -
The odd field images up to a n-1 are stored in the odd field memory 7, and the even field images b2 to bn in the figure are stored in the even field memory 8, respectively. How the images stored in each of these memories 7.8 are read out is determined by the MODE signal. That is, when the MODE signal is set to 1'', the first selector 9 can be activated, and an image double read out every horizontal scan can be selected from the odd field memory 7. MOD E signal “o”
, the second selector 11 can be activated via the inverter 10, and the images read out from the odd field memory 7 and the even field memory 8 and alternately selected by the multiplexer 12 every horizontal scan. is selected. The selectors 9 and 11 first output the character image to the area detection circuit 13. The area detection circuit 13 detects the area of the character line 2 from the input character image, and sets the starting point address and area information in the address control circuit 6. The address control circuit 6 reads out the image from the odd field memory 7 or the even field memory 8 again.
At this time, only the detected character area is read out. The image of the character area is input to the multiplexer 14 via the selector 9 or 10. ? Lunaplexer 14 selects either the character image provided via selector 9 or the character image provided via selector 11 based on the MOD IEE signal, and outputs the selected character image to character recognition circuit 15 . The character recognition circuit 15 performs character-by-character detection, extraction, sampling, and calculation of similarity with the comparison pattern 16, and outputs recognition results.

In the above configuration, if the input character image [1] is a still image as shown in FIG. The image data D1 and the image data D2 stored in the even field memory 8 are read out alternately and an assembled character image DO is subjected to recognition processing.

On the other hand, as shown in FIG. 4, when the input character image fl+Di is a moving character image, the image data D1 of the odd field and the image data D2 of the even field are shifted, which may affect recognition accuracy. is possible. In this case, the output character image Do' obtained by double scanning the odd field image data D1 expresses its characteristics more sharply than the output character image IDO obtained by reading out the images of each field alternately. There is. Therefore, the MODE signal is set to "1" so that the double readout image of the odd field memory 7 is subjected to recognition processing.

In this device, the peripheral circuits of each field memory and their control method will be described in more detail below.

That is, FIG. 5 is a diagram showing the configuration of the peripheral circuit.

XW is a write pulse for counting up the X counter 22 via the gate 2) and writing binary image data to the odd field memory 7 and the even field memory 8. XW consists of an odd field component and an even field component, and VBP is a pulse generated during the vertical retrace period when the odd field component changes to an even field component (see FIG. 6). This VB
P counts up the ΔY counter 24 through the gate 23.

YW is a pulse that counts up the Y counter 27 via the gates 25 and 26 during writing, and when XW is 51
One pulse is generated every two pulses generated.

Writing of the binarized image data into the memory 7.8 is performed as follows. Due to the occurrence of CLR, the X counter 22,
The outputs of ΔY counter 24 and Y counter 27 are all "o"
”. Therefore, the output from the ΔY counter 24 Δ
YQ becomes 110 I+, and odd field memory 7 is selected. As a result, a 512×256 odd field pattern is written into the odd field memory 7 by the generation of xW and YW pulses.

Next, the output ΔYQ of the ΔY month counter 4 is counted up by VBP and becomes ll 1 t+, so the even field memory 8 is selected by the inverter 28.

As a result, the next generated 512 x 256 even field pattern is set to the even field memory 8 by xw and yw.
will be written in.

Next, the case of reading out the binary image data stored in the memory 7.8 will be explained.

XR is a pulse that counts up the X counter 22 when reading from the memory. YRl is ΔY month counter 4,
Further, YF3 is a pulse that counts up the Y counter 27, and both are generated for each XR scanning line. By inputting YRl, the ΔY month counter 4 alternately changes to 1'' and 11 ON, so that the data from the odd field memory 7 and the even field memory 8 are alternately output.

On the other hand, in order to double read the odd field memory 7, the output of the ΔY month counter 4 is fixed at "0" by the CLR. Next, if the read pulse of YF3 is frequency-divided by 1/2 by the F/F 29 and applied as a clock to the Y counter 27, the same scanning line of the odd field memory 7 will be scanned twice.

This results in double reading.

In this configuration, YRl and YF3 are the MODFs mentioned above.
It becomes a signal.

As described above, according to this embodiment, even when a character image is moving, accurate recognition is possible by double reading of odd field images. In this case, since there is almost no difference between the still image and the moving side, there is no need to prepare separate comparison patterns for both, and there is no need for movement correction.

Of course, since a high-speed shutter is not required, the recognition performance can be improved while maintaining the low cost of the device.

Note that the present invention is not limited to the embodiments described above.

For example, it is optional whether to double read odd fields or even fields, and it is also possible to compare the recognition results of double read images of both odd and even fields and adopt the answer with a high degree of similarity. . It is also possible to use odd and even field memories in a vibrating manner.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a character reading device according to an embodiment of the present invention, FIG. 2 is a diagram for explaining the interlace scanning method of a TV camera connected to the device, and FIG. A diagram showing how the device generates an output image for recognition from an input image of a static character image, and FIG. 4 is a diagram showing how an output image for recognition is generated from an input image of a moving character image. , FIG. 5 is a block diagram showing further details of the memory surrounding the device, and FIG. 6 is a timing chart of various pulses to explain the operation of the circuit. 1...Product, 2...Character line, 3...TV camera, Di...Input character image, Dl...Image data stored in odd field memory, D2...Even field memory Image data stored in , [)0... Output character image formed by reading odd and even fields alternately, [) o l... Output formed by double reading odd fields. Text image.

Claims (2)

[Claims] (1) An imaging device that performs interlaced scanning, an odd field memory that stores odd field character images obtained by imaging the object to be read with the imaging device, and an odd field memory that stores character images of odd fields obtained by imaging the object to be read by the imaging device; means for alternately reading character images from the odd field memory and the even field memory for each scan; means for reading a character image in an even field memory by double scanning; means for selecting the alternately reading means and the means for reading by double scanning according to a moving state of an object to be read; A character reading device comprising: means for detecting a character area of a read character image; and means for character recognition of the character image detected by the means. (2) A comparison pattern for character recognition of character images read out alternately and a comparison pattern for character recognition of character images read out by double scanning are the same pattern. The character reading device according to item 1.