JPS5815829B2 - Character/figure recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character/figure recognition system that allows highly efficient recognition of characters/figures, particularly freely drawn handwritten characters.

Handwritten characters produce various patterns depending on the person who draws the same character, and cannot be recognized as easily as characters drawn using a machine that have a fixed pattern.

Conventionally, a method for optically reading and recognizing handwritten characters is to break down a character pattern into several parts, for example in the horizontal direction, and assign the characteristics of the pattern belonging to each part to one of a preset feature group. Depending on the classification, it is converted into a simple symbol, and the combination of symbols obtained from each part is recognized by comparing it with a dictionary pre-set in the central processing unit by processing by an electronic computer.

Here, the dictionary provided in the central processing unit is written in advance in order from symbol combinations with a high probability of appearance to combinations with a very low probability of appearance. It is written that it corresponds to

However, since handwritten characters are written by various people in various styles, a dictionary of enormous size is required to ensure accurate recognition regardless of the character pattern, and it has not been put into practical use as a recognition method for character reading devices. There was still a problem in providing it.

By the way, when we examine the combinations of character patterns and symbols representing characteristics, we find that even if the character patterns have similar characteristics, the positions where the characteristics appear are diverse.

To specifically describe the original pattern "+", Figure 1a
, b, c, and d are all very similar to the original pattern, but the positions where the features appear vary.

That is, each pattern has been horizontally decomposed into eight parts, and now the characteristics of the pattern can be expressed as the sections /A and /B in Figure 2.
, /C, symbol N indicates that a black region crosses the section vertically, symbol B indicates that there is a black region that crosses within the section, and symbol C indicates that there is no black region within the section. If the patterns are set and classified as follows, the patterns shown in FIG.

Although the illustrated pattern is just an example, it is easy to imagine that even topologically similar patterns can have a considerable number of combinations of symbols due to positional diversity, and even more so when reading ordinary characters. It becomes necessary to classify and convert symbols into many types of characteristics, and more complex combinations of symbols appear, requiring an extremely large capacity as a dictionary.

As the dictionary becomes larger, not only does the area occupied by the dictionary on the central processing unit increase, but when the contents are referenced serially, the time required for reference becomes longer in proportion to the size of the dictionary, which reduces the recognition operation time. Shortening is impeded, and devices with an order of magnitude greater capacity are required to make reference possible.

The present invention has been made in view of the above-mentioned problems of the conventional method, and instead of immediately comparing the symbol combinations of the decomposed character parts with the dictionary in the central processing unit as in the conventional method, the characteristics of the pattern are A simple method that focuses on the positional diversity that appears, breaks down a range of characters into segments in a certain direction, converts the features of each segment into symbols in the order of the intercepts, and expresses the features by combining them with the symbols of the previous intercept. Once converted into a pseudo pattern, the pseudo pattern is transitioned according to the characteristics of the intercepts obtained sequentially to form the final pseudo pattern, and the pseudo pattern is compared with a dictionary to improve the efficiency and accuracy of the operation. It is something that

Next, the present invention will be explained in detail with reference to Examples.

Figure 3 is a block diagram showing the outline of the character/figure reading device.
The character/figure written on the form is read by the optical system 2 and converted into an electric signal by the photoelectric conversion circuit 3, and then converted into a digital signal by the A/D conversion circuit 4. For example, the digital signal corresponding to one character is After sampling and performing pre-processing such as noise processing, the data is stored in the storage area 500 in the storage device 5.
It is stored in a two-dimensional distribution state with almost the same pattern as the original pattern drawn on the form surface.

For example, if an original pattern "+" is drawn on the form surface and this is read by the optical system 2, an electrical signal is stored in the storage area in a pattern similar to the original pattern "+".

As shown in FIG. 4, the storage area 500 is made up of, for example, 32 shift registers each having a capacity of 24 bits connected in a row, and a two-dimensionally read pattern is stored in the capacity of 24×32 bits.

Here, each column of the shift registers 501 to 532 corresponds to one segment obtained by horizontally decomposing the range of one sentence.

The black area in which the pattern of each section is drawn is classified according to preset features as described later, and converted into symbols representing each feature, and each symbol is then converted into a unique binary code suitable for signal processing. converted into a code signal.

Conversion to a binary code signal is performed using PLA (Progra
This is performed by comparing preset features with the black area of the drawn section using an encoder 533 consisting of a Logic Array (m Logic Array) or the like.

The conversion into symbols is performed for each segment of the character/figure, and the resulting binary code signal is transferred to the recognition circuit 6 at the next stage.

In the above embodiment of the present invention, when a signal for recognition is transferred from the storage device 5 to the recognition circuit 6 via the encoder, the shift register 5 at the bottom stage of the storage area 500 is
01 to the contents stored in the top shift register 532 in units of 24-bit shift registers, and the symbols of each slice are serially derived and supplied to the recognition circuit 6.

A central processing unit 7 described below is connected to the recognition circuit 6.
Based on the relationship between the symbol signal and the pattern written in the central processing unit 7, a corresponding pseudo pattern is read out and pattern information is output to the output device 8.

The central processing unit 7 is programmed in advance with a process in which patterns assumed by the symbols change based on the symbols obtained for each slice.

That is, through computer processing, a transition diagram is obtained by assuming what combinations of symbols will appear in various patterns in advance, and the relationship between the symbol combinations obtained from the transition diagram and the pattern represented by it is calculated. is programmed, and a pattern is read out corresponding to the symbol given to the recognition circuit 6.

In the present invention, a symbol is obtained for each slice, so depending on the symbol of the initial slice, a simple pattern is generally read out, and this pattern passes through transitions depending on the symbols given sequentially.

Therefore, the program of the central processing unit 7 is written with a process of transitioning from a simple pattern to a complex pattern corresponding to the characteristics of the symbols given sequentially, and the final pattern is derived during the transition.

All of the above-mentioned patterns derived from the central processing unit 7 are pseudo patterns, and the final pseudo pattern is compared with a dictionary provided in the central processing unit 7 in the output device 8 to form the final character figure. Recognition takes place.

In the dictionary, at least what kind of character/figure the finally obtained pseudo pattern corresponds to is written.

Focusing on the positional features of character shapes, we derive the final pseudo pattern by fleshing out the simple pseudo pattern with the next given feature as described above, and then derive the final pseudo pattern, and then compare this final pseudo pattern with a dictionary. By determining the character/figure, unnecessary verification can be omitted and recognition operation can be made more efficient.

In order to simplify the explanation, the recognition method of the present invention will be explained in more detail by citing the case where the original pattern "+" is recognized.

The "+" pattern read by the optical system 2 is stored in the storage area 500 of the storage device 5 via the photoelectric conversion circuit 3 and the A/D conversion circuit 4.

The contents of the storage area 500 are stored in the second shift register column for each shift register column.
Symbol i set in advance in light of the features classified in the diagram
The signal is converted into either B or C and output as a binary code signal set corresponding to each symbol.

On the other hand, in the storage device of the central processing unit 7, a pseudo pattern obtained by combining symbols that appear in advance through electronic computer processing is written as shown in FIG.

Therefore, when the contents of the storage area 500 are sequentially applied from the bottom of the shift register to the recognition circuit 6 as symbol signals via the encoder, the symbol C is outputted first, and then a pseudo pattern "S" indicating a start is outputted.

When the symbol A is subsequently supplied to the recognition circuit 6 while the pseudo pattern "S" is being output, it is compared with the content written in the central processing unit 7 and the symbol A is displayed as 1A-8 in FIG.
First, the simple pseudo pattern written at the position corresponding to the intersection of j is determined to be "1", and the content given to the output device 8 changes to "1".

In this state, when the next symbol B is supplied to the recognition circuit 6, it is again compared with the contents of the central processing unit 7 and is shown in FIG.
It is determined that the pseudo pattern written at the intersection of 1-BJ is "T", and the content given to the output device 8 transitions to the pseudo pattern rTJ.

Similarly, based on the symbol signal given to each slice, comparison with a pseudo pattern previously provided in the central processing unit 7 is continued, and finally a pseudo pattern "+" is output.

The final pseudo pattern is further compared with a dictionary in the central processing unit, and it is determined that the original pattern is "+".

FIG. 6 is a transition diagram showing the above judgment operation from the start.

The above practical example explained the recognition operation of a pattern that can be expressed by three types of extremely simple features, but in order to perform practical recognition of characters and figures, it is necessary to set a large number of symbols in advance. Furthermore, the amount of pseudo patterns that are expected to appear corresponding to the set symbols also increases.

For example, when recognizing 10 types of characters such as 0, 1°, 2, ... 9 that have been conventionally used for OCR, first, the drawn character area is separated into one horizontal direction, and the As shown in FIG. 7, 16 types of features are set as a pattern recorded on each separated section depending on the width, number, etc. of the black area, and the set features are shown in the figure.
A unique symbol is given to each as shown in .

Therefore, the character pattern stored in the storage area is expressed by a combination selected from 16 types of symbols depending on the status of the black area of each section, and the symbols are converted for each section and transferred to the recognition circuit 6.

On the other hand, in the central processing unit 7, pseudo patterns assumed in combination with symbols corresponding to the 16 types of symbols are written in advance, and pseudo patterns corresponding to the symbols supplied to the recognition circuit 6 are read out. Output.

The read pseudo-pattern is further modified with the next given symbol, and the pseudo-pattern is outputted through transitions that take into account the characteristics of the symbols.

8A and B show the above 16 written in the central processing unit 7.
This diagram schematically shows the relationship between type symbols and pseudo pattern transitions.

For example, when determining the original pattern "8" with the transition diagrams shown in FIGS. 8A and 8B programmed into the central processing unit 7, the symbol "H" is first determined by the black area "-" at the bottom. is obtained, and from FIG. 8, a pseudo pattern "-" (■) corresponding to rHJ is output.

Next, the symbol "HL" is obtained by the black area of "L", and the pseudo pattern transitions from "-" to rLJ■.

Then, the black area “-” in the center makes the symbol “H” appear again.
” is obtained, the pseudo pattern transitions to “ko” ■, and with the given symbol “RH” and symbol “H”, the i-like pattern transitions from “pi” ■ to “self” ■, and the final The pseudo pattern is "2" and is determined to be "2" after checking with the dictionary.

FIG. 9 shows that when characters O to 9 are recognized based on the above transition diagram written in the central processing unit, the final pseudo pattern is obtained by sequentially transitioning from a simple pseudo pattern, and this final pseudo pattern is 12 is a flowchart showing what character is determined by comparing a pattern with a dictionary.

In all of the above practical examples, the contents stored in the storage device as the range of a literary society are derived from the bottom row, but by changing the symbol conversion circuit and transition diagram, the range of a literary society is derived from the top row. Similarly, it can be derived from either the right or left.

As described above, according to the present invention, recorded patterns are sequentially derived by decomposing the read pattern in a predetermined range into sections in the f direction,
The derived reading pattern of each section is converted into a unique symbol in light of preset characteristics, and each of the above is converted from a central processing unit that has previously written a pseudo pattern that is assumed to be expressed in correspondence with a combination of symbols. A pseudo pattern corresponding to each symbol of the section is read out and the pseudo pattern is sequentially transitioned to form a final pseudo pattern to be recognized, and the final pseudo pattern is checked against a dictionary to recognize characters and figures. Therefore, character shape feature extraction is extremely difficult because it is first converted into a simple pseudo-pattern with the features of each intercept, and then the previous pseudo-pattern is transitioned with the features given one after another to obtain a more complex final pseudo-pattern. This is performed efficiently and can appropriately respond to positional deviations of features.

In addition, in the comparison operation with the dictionary, since the final pseudo pattern is judged, it is possible to eliminate unnecessary operations compared to the case of simple comparison as in the conventional method, and to speed up the recognition operation. In order to recognize characters and figures to the same degree as the above, the capacity required of the central processing unit is smaller than that of the conventional system, and a practical device can be obtained.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the appearance form of character figures, FIG. 2 is a diagram showing the drawn characteristics of the character figures shown in FIG. 1, and FIG. 3 is a block diagram of an OCR device embodying the present invention. FIG. 4 is a diagram showing the main parts of the device, FIG. 5 is a diagram showing the relationship between symbols and pseudo patterns used to explain the operation of the present invention, and FIG. 6 is a transition diagram showing an embodiment of the present invention. FIG. 7 is a diagram showing the extracted features and symbols of the numerical pattern used to explain the operation of the present invention, FIG. 8 is a diagram showing the relationship between the combinations of symbols in FIG. 7 and pseudo patterns, and FIG. 3 is a flowchart showing a process of recognizing a number pattern according to the invention. 5...Storage device, 500...Storage area, 501-532...Shift register, 6...
...Recognition circuit, 7...Central processing unit, 8.
...Output device.

Claims (1)

[Claims] 1 A character figure is decomposed into a plurality of sections in a certain direction, optically read and converted into electrical signals, and the converted electric signals in a predetermined range are made to correspond to the distribution of the character figure and are two-dimensionally stored in a storage device. The content of each section of the storage device is derived one-dimensionally from the end, and the features of the read character/figure are classified in correspondence with the features of the figure set in advance, and The classified features are compared with the pseudo-patterns written in the central processing unit in correspondence, the corresponding pseudo-patterns are read out, and the read-out pseudo-patterns are sequentially transitioned for each section to be recognized. A character/figure recognition method characterized in that a final pseudo pattern is formed, and recognition is performed by comparing the final pseudo pattern with a dictionary in a storage device.