JPS5897784A - Character discriminator - Google Patents

Abstract

PURPOSE:To simplify a character discriminator, by calculating the degree of resemblance by using the line edge of a character to the discriminating information and then discriminating characters. CONSTITUTION:The information obtained by tracing the line edge of a character is converted to obtain a 1-dimensional code string of each direction through a means 1. Such means 1 is provided with a direction coding circuit 4, a sorting circuit 5, a gate 6, a feature point extracting circuit 7, a photoelectric converting circuit 8, a synchronizing signal producing circuit 9 and a region signal producing circuit 10. The 1-dimensional code string is compressed or expanded by a normalizing circuit 2 and set at a certain length. The output of the circuit 2 is fed to a dictionary and a resemblance discriminating circuit 3. Then the 1-dimensional code string of a stored standard pattern is compared with a 1-dimensional code string which is converted into a certain length. Thus the characters are discriminated.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a character identification device such as an optical character reading device.

In an optical character reading method, the character pattern is generally narrowed down, the center line is determined, etc., and the pattern is converted into a line shape, and then the pattern is converted into a line shape, and a code string such as an 8-way code is used, or the character pattern is A method is used to recognize characters by extracting partial features from the character, expressing them in a feature code string, matching them with pre-coded nine standard patterns, or passing them through a sequential logic circuit.

In particular, character identification methods using similarity normalize character patterns to create patterns with uniform size and compressed information, and then calculate the similarity between this and standard patterns held in a dictionary. I am giving an answer.

Furthermore, this method has the advantage that it is easy to create a dictionary due to advances in memory and the like.

By the way, some of the features that represent characters include those that do not use all of the elements that make up a character, the ``black'' part.

For example, the fineness of a character (the outline of a character pattern) is known to relatively well indicate character characteristics, and has been used in many identification methods in recent years.

However, it is also known that character identification by feature extraction using this edge information requires a great deal of effort because the amount of information is large, although the information processing is simple.
The present invention has been made based on the above-mentioned circumstances, and is characterized by the ease of dictionary creation and high accuracy of the similarity method, and character recognition that takes advantage of the characteristics (easiness of handling of the primitive Shibaturn extraction method). The purpose is to provide equipment.

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

FIG. 1 shows one embodiment of the present invention.

Here, as shown in Figure 1 (1), the means to convert the information obtained by tracing the edges of the characters and obtain one-dimensional codes in each direction is to convert the information obtained by tracing the edges of the characters into directional codes. It consists of a conversion circuit 4, a classification circuit 5, a gate 6, a feature point extraction circuit 7, a photoelectric conversion circuit 8, a synchronization signal generation circuit 9, and a region signal generation circuit 10.

The direction encoding circuit 4 is connected to a classification circuit 5, and the data classification circuit 5 is connected to a gate 6, which converts codes classified into 8 or 16 directions into binary data (3 bits for 8 directions, 4 pits for 16 directions). 8 or 16 signals with detected feature points.
It is classified as

The gate 6 is connected to a feature point detection circuit 7 and a region signal generation circuit 10.
, and outputs the signal received from the feature point detection circuit 7 in synchronization with the signal from the area signal generation circuit 10.

The feature point detection circuit 7 is connected to the photoelectric conversion circuit 8, and the signal sent from this circuit detects the partial feature of the edge.

The photoelectric conversion circuit 8 is connected to the synchronization signal circuit 9 and converts the character pattern into a digital signal block by block using COD or the like.

The synchronizing signal generating circuit 9 is connected to the area signal generating circuit 1o, and generates a signal for the photoelectric conversion circuit 8 to scan over a character.

The area signal generation circuit 10 is connected to the synchronization signal generation (b) path 9, receives this signal, and generates a signal for dividing data into blocks.

Next, the means for compressing or expanding the one-dimensional code string to a constant length includes a shift register 11, a shift register 12, a comparator 1 node 13, a counter 14, and a gate, as shown in FIG. It consists of 15.

The shift register 11 is connected to the direction encoding circuit 4, and takes in, latches, and shifts binary data obtained therefrom in synchronization with a shift clock (not shown).

The shift register 12 is connected to the shift register 11, and takes in, latches, and shifts a signal obtained from the shift register 11.

Comparator 1/13 is connected to shift register 11 and shift register 12, and compares their output signals.

The counter 14 is connected to the comparators 1 and 13 and counts the output of the comparators 1 and 13 to determine the length of the one-dimensional code string.

The gate 15 includes a comparator 13 and a shift register 12.
The output of the shift register 12 is sent to the buffer 17 according to the output signal of the comparator 13.

Next, a means for storing standard patterns of one-dimensional code strings possessed by various characters is compared with the one-dimensional code string of the standard pattern obtained by the means and the one-dimensional code string converted to a certain length, and the one-dimensional code string is identified. As shown in No. 9 (3), the means for
It consists of

c p U16ti, buffer 17 via Hass 19;
It controls the memory 18 and the like.

Buffer 17 is connected to gate 15 and temporarily holds the output from the shift register.

The memory 18 stores standard pattern code strings in advance and outputs them in response to instructions from the CPU 16.

The above is the configuration of one embodiment of the present invention.

Next, the operation will be explained with reference to the drawings.

For example, when identifying the character "2" as shown in FIG. Ru.

■ Based on the predetermined starting point ■, 8 as shown in Figure 3 (a).
The line fibers are detected in the direction and tracked block by block.6 First, since there are 8 directions, the direction of the line fibers is directly above, diagonally upper right, right, diagonally lower right, directly below, diagonally lower left, and left. , are classified diagonally on the upper left, but here they are classified into 1, 2. 3. 4. 5 in order.
．． Let's number it 6.7°8. Also, if each of these directions is expressed using 3 binary bits, for example, roo
oJrOoljroloJrollJrloooJrlO
It becomes like IJrllOJrllllJ.

Next, regarding the character "2" shown in Figure 2, considering the direction of the wire using the direction code and 3 binary bits, we can define the edge direction of the starting point as direction 3 (Tsumushi, roloJ in binary). Then, next is 3 (010), next is ryo4 (011), next is [ol
l, next #'15 (100), etc., the result is,
If this code string is fS, then fs=3344
5666655433355777777111112
22287677222, and if the length of the code string is t, then t=40.

■ There are two ways to normalize a code string: compression and expansion.If we consider compression here, the edge direction in a code string is, if the blocks that divide characters are small enough, Paying attention to the fact that they are continuous, these continuous codes can be treated as one code.

(2) The binary data sent from the direction encoding circuit 4 is latched into the shift register 11 in synchronization with the shift clock.

(2) The data latched, shifted and output by the shift register 11 is again input to the cascade-connected shift register 12 and shifted.

■ Data is 3 bits in binary and shift register 11°12
If the number of 0 bits is 3, the output of shift register 11 and the output of shift register 12 connected in cascade will necessarily differ by 3 bits.

For example, if fS=33445666--- is input 77 times in order, the first "3・3" 1
Paying attention to -, and expressing this in binary, it becomes "010.010".

However, since the 3-bit shift registers are connected in cascade, the beginning of the code pit that entered earlier is output from the shift register 12 in the lower stage, and at the same time the beginning of the code pit that entered later with a 3-bit delay is output. It is output from the upper-stage sister register 11. As a result, the outputs from the two shift registers are 0, 011, 1.0, and O.
Totoru.

(2) These outputs (0 and 0, 1 and 1, O and O) are added to the input of the comparator, but even when these are compared, the output of the comparator is "0".

(2) The counter 14 counts the output of the comparator 13 to determine the length t of the code string, but in this case it does not count because there is no output.

(2) If there is no output from the comparator 13, there is no output from the gate, so no data is output to the buffer 17.

As described above, if the direction code is "33" two times in a row, it will be normalized to "3". This length is also a predetermined value. If the resulting output is f, then f,=345665433
577711122876722, so t=24.

Also, if the direction code is different, for example "2/6",
These are "010.101" in binary, and each input to the comparator is different for 0, 1.1, 0.0, and 1, so the comparator 13 outputs an output and the counter updates the count. However, data is also output to the buffer 17.

The above description assumes that all the bits of the code are different, but if only one bit or two bits are different, the same process can be performed by performing an appropriate logical operation.

In addition, when expanding code strings, these strings are left-justified and a code with a meaning is added as a missing code in order to make the string a predetermined length using a counter.

■ The normalized code string is held in the buffer 17. For this buffer 17, approximately 128 words is sufficient for alphanumeric characters and kana symbols.

■ To identify characters consisting of these code strings, under the control of the CPU 16, standard character pattern code strings, commonly called a dictionary, are stored in memory (codes transferred from other storage media). If string data is held, read the code string. Naturally, this is not limited to just one.

(2) The CPU 16 compares these read code strings with the code string normalized from the input data, and selects the most probable one as the answer.

This can be done, for example, by storing these code string data and standard pattern code string data in each register (not shown) of the CPU 16, performing logical operations, adding them, and checking the flag. Just do it.

Also, by using the functions of the CPU 16, the two shift registers, counters, and gates are omitted and the CIPU 16 is used.
The normalized code string may be directly transferred to the buffer 17 and processed using software.

■ The identification result is output to the outside under the control of the CPU 16.

The above is the operation of the embodiment of the present invention.

The effects of the present invention are as follows.

Since the automation of dictionary creation that the similarity method has can be used, it is possible to create a highly accurate dictionary in a short period of time.

If the pattern to be matched is only the character line edge (periphery) information, the amount of information is compressed (for example, a code length of 40 is reduced to 24), and handling is easy. Therefore, the pattern processing time can be shortened, and the required memory is also reduced, so that the amount of hardware is reduced.

As described above, the present invention uses line fibers of characters as identification information,
This method is characterized by simplifying the device by calculating the degree of similarity and identifying characters.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a plan view showing a procedure for tracing line fibers of characters, and FIG. 3 is a plan view showing specific classifications of line fibers. 1... Primitive feature extraction circuit 2... Normalization circuit 3.
... Dictionary and similarity identification circuit 4 ... Direction coding circuit 5 ... Classification circuit 6 ...
・Gate 7...Feature point extraction circuit 8...
Photoelectric conversion circuit 9...Synchronization signal generation circuit 10...
・Area signal generation circuit 11... shift) L/register 2
...Shift register 13...Comparator 14
...Counter 15...Gate 16...
CPU 17...Buffer 18...
Memory 19...Pass agent Patent attorney Noriyuki Chika (and 1 other person)

Claims (1)

What is claimed is: means for converting edge features of a character into a one-dimensional code string of directional codes by tracing the fineness of the character; and compressing or expanding the one-dimensional code string obtained by the means; means for storing standard patterns of one-dimensional code strings of various characters; one-dimensional code strings of standard patterns obtained by the means; and one-dimensional code strings obtained by the means for making the length constant. A character identification device comprising: means for comparing and identifying dimensional code strings.