JPH0198094A - Character recognizing circuit - Google Patents

Abstract

PURPOSE:To determine candidate characters automatically at high speed by storing distance data obtained by comparing a character to be recognized with dictionary characters, comparing read distance data with a threshold level, and regarding data smaller than the threshold level as output candidate characters. CONSTITUTION:The distance data DIST 01-11 obtained by comparing the character to be recognized with the dictionary characters are inputted to a multiplexer (MPX) 12. A histogram generation part 34 writes the frequency of generation of the distance data in a histogram area of a memory 18 by using the distance data as addresses. A threshold level determination part 36 reads the memory 18 in increasing address order to integrate histograms in respective address and quits reading the memory when the output specified number OTD of data is arrived, thereby using the current address as the threshold level. A data output decision art 38 compares the distance data read out of the memory 18 with the threshold value from the determination part 36 and decide data smaller than the threshold level as output candidate characters. Thus, candidate character determination processing is performed automatically at high speed.

Description

[Detailed Description of the Invention] [Summary] The present invention relates to a character recognition circuit that compares read characters with dictionary characters, selects a predetermined number of similar characters, and uses them for recognition. A character recognition circuit that compares the target character with dictionary characters and selects dictionary characters with a predetermined number of short distances as candidate characters, with the aim of outputting candidates quickly and automatically by using hardware , a histogram creation unit that receives distance data obtained by comparing a recognition target character with a dictionary character and writes the distance data as an address and the number of occurrences of the distance data as data in a histogram area of the memory; The histogram of each address is read in descending order, the reading is stopped when the specified number of output data is reached, and the threshold value determination circuit that uses the address at this time as the threshold value is used to read out the distance data area of the memory and read out the histogram of each address. and a data output determination unit that compares the distance data and the threshold value left, and selects a character smaller than the threshold value as an output candidate character.

[Industrial application field]

The present invention relates to a character recognition circuit that compares read characters with dictionary characters and selects a predetermined number of similar characters for recognition.

[Conventional technology]

When recognizing characters with many types, such as kanji, compare the target character with all characters in the dictionary, create a histogram of the comparison results (distance data), and use the histogram to recognize only the preset number of characters. A process is performed in which a distance threshold is determined so that candidate characters can be output, the threshold is compared with distance data, and dictionary pointer data and distance data are output even if the distance is smaller than the threshold.

In other words, the method is to first narrow down the candidates and then select the appropriate one from the narrowed down candidates.

[Problem that the invention seeks to solve]

Conventionally, the series of processes for selecting candidates described above has been performed by software, which has the drawback of being extremely time consuming.

It is an object of the present invention to improve this point and to perform all of the histogram creation, threshold value determination, and output determination processing in hardware, thereby making it possible to output candidates quickly and automatically.

[Means for solving problems]

The above purpose is to input distance data obtained by comparing the recognition target character with the dictionary characters in a character recognition circuit that compares the recognition target character with dictionary characters and uses dictionary characters with a predetermined number of short distances as candidate characters, Histogram creation unit (34) writes the distance data as an address and the number of occurrences of the distance data as data in the histogram area of the memory (18).
) and a threshold determining circuit (36 ), and a data output determination unit (38) that reads the distance data area of the memory, compares the read distance data with the threshold value, and selects characters smaller than the threshold value as output candidate characters. achieved.

[Effect]

According to this configuration, since histogram creation, threshold value determination, and data output judgment can be performed using hardware, it is faster and faster than the conventional software-based method.
Automatic processing becomes possible and is extremely effective.

〔Example〕

FIG. 1 shows an embodiment of the present invention. 12 is a multiplexer, 14.16 is a counter, and 18 is a local memory (SR
AM), 20. 22 is an adder, 24.degree. 26 is a register, 28.30 is a comparator, and 32 is a multiplexer. These adder (+1 circuit) 20, register 24, and multiplexer 32 create a histogram creation section 34,
Further, the adder 22, the register 26, and the comparator 28 function as the threshold value determining section 36, and the comparator 30 functions as the data output determining section 3.
8. Although not shown, a circuit for controlling all of these is also provided.

FIG. 2 shows the configuration of the local memory 18. As shown in the figure, the local memory 18 has an output pointer data area (
4 bytes), output distance data area (4 bytes),
There is a histogram area (16 bytes), a distance data area (32 bytes), and an output work area (32 bytes).

In the circuit shown in FIG. 1, histogram creation, threshold value determination, and data output are executed sequentially in this order (however, before the first histogram creation, the histogram area of the local memory 18 is cleared. It is shown in Figure 3. In Figure 3, *CLK is the system clock, *HMC
ST is a clear start signal, which enters the counter 14 through the OR gate G2 and clears it. Then, at the rising edge of *HMCST, a clear signal HSTCLR indicating "clearing" is generated, which passes through the OR gate G1 and enters the multiplexer 12, and is sent to the distance data DIS.
Switch from the TOI~11 side to the counter 14 side. The counter 14 receives a clock input after the rising edge of H5TCLR*
Count HCLK2 and set the count value to 0.1, 2.・・・・・・
I'm going to change it to... This passes through multiplexer 12,
It enters the local memory 18 through the local memory address bus AB and becomes the write address of the memory 18.

O is manually input to the memory 18 as write data, and this is written to addresses 0, 1, 2, . ......, 40
95, and thus the histogram area is cleared. C8 in FIG. 3 is a chip select signal, and WE is a write enable signal.

When 0 is written to all addresses in the histogram area, address end 80ED rises, and at the falling edge, clear end * MMCHD is generated, and at this falling edge, the clear signal H
STCLR falls. *HMCHD becomes the distance calculation start signal.

The histogram is created by repeating the process of increasing the data in the histogram area by 1, using 1/2 of the distance to the contents (characters) of the dictionary as an address. A time chart of this process is shown in FIG. Distance calculation is done outside this circuit, distance data old 5TO1~
11 to the multiplexer 12 of this circuit,
It is written to the distance data area of the local memo I718 together with the dictionary pointer data to the output work area. If the number of characters in the dictionary is N, there are N pieces of distance data and pointer data for each character to be recognized, but the histogram is created each time one piece of distance data and pointer data is written to the distance data area and output work area. Let's do it. *HISTST in Fig. 4 is a write start signal, and at its rising edge, *HIST DRUP signal is generated, and signal C
3, OE (out enable).

When WE is generated, memory reading and writing are performed, and one cycle ends with an end signal *FIISTED.

*HISTED instructs the start of the next distance calculation.

When creating a histogram, the multiplexer 12 uses distance data D
IST side, so that DIST passes through multiplexer 12 and address bus AB into local memory 18 and becomes the read address of the memory. The read data passes through the data bus DB, enters the +1 circuit 20, is incremented by 1, is launched into the register 24, passes through the multiplexer 32, reaches the data bus DB, and is written to the above address in the local memory 18. Histogram creation is performed by repeating this process.

After creating histogram, start signal TH5T for determining threshold value
Threshold value determination processing is performed. This time chart is shown in FIG. When the start signal THST is output, the threshold signal TH rises at its rising edge, enters the multiplexer 12 through the OR gate G1, and switches it to the counter 14 side. The counter 14 is cleared at TH5T and counts when the clock *HCLK2 is input. The count value of the counter 14 becomes the address of the local memory 18, and the histogram of each address (distance) is read from the memory. In the histogram area, the distance data is used as an address and the number of occurrences of the distance data is written, so there is a histogram of a small distance at the smaller address. Therefore, if you repeat the process of reading from the beginning of the histogram area (address O), inputting the read data to the adder 22, adding it to the contents of the register 26, and moving the addition result to the register 26, the register 26 will contain The cumulative value of the histogram for each distance from the smallest distance is obtained. This (A) is compared with the output designated data number OTD (B), and when it reaches A2B, the signal THRD is outputted to stop reading the local memory. The memory address at which reading is stopped becomes the threshold for retrieving the planned number of candidates (in detail, since the address is 1/2 of the distance data as described above, it is twice the address at which reading was stopped. is the threshold). This is the end signal of this process * THE
D becomes a start signal for data output processing, which is the next process.

FIG. 6 shows a time chart of data output processing. *
The OUT cycle rises at the rise of THED, and IC3,
*OE performs memory reading, and *WE performs memory writing. First, the distance data area of the memory 18 is read out (this read address is generated by the counter 16 that counts the clock *HCLK3),
A comparator 30 compares the read data (B) with a threshold value (Bi, which is generated by the counter 14) and produces an output OUT if A>B. When the output OUT appears, the distance data and dictionary pointer are stored in the memory 18 as shown in process (II).
Write to the output distance data area and output pointer data area. If the output OUT is not output, only memory reading is performed as in process (I), and writing is not performed.

The above process (
1) When (II) is performed, the data end DTED rises as shown by process (n[), the OUT cycle falls, and the end signal *0UTED is output to notify the end of the process.

The candidate characters are output by reading out the area written in step (n). The number of candidate characters to be output is determined by the output designated data number OTD, but it is not a fixed number because it is a unit of distance.

〔Effect of the invention〕

As explained above, in the present invention, histogram creation, threshold value determination, and data output determination are performed by hardware (processing-dedicated LSI), so unlike conventional methods, it does not take much time (i.e., Decision processing can be performed automatically and quickly.

[Brief Description of the Drawings] Figure 1 is a block diagram showing the configuration of the circuit of the present invention, Figure 2 is a block diagram showing the configuration of local memory, Figure 3 is a clearing time chart of the histogram area, and Figure 4 is the histogram. FIG. 5 is a creation time chart, FIG. 5 is a threshold value determination time chart, and FIG. 6 is a data output determination time chart. In FIG. 1, DIST is distance data, 18 is a memory, 34 is a histogram creation section, 36 is a threshold value determination section, and 38 is a data output determination section.

Claims (1)

[Claims] In a character recognition circuit that compares a recognition target character with dictionary characters and uses dictionary characters with a predetermined number of short distances as candidate characters, distance data obtained by comparing the recognition target character with the dictionary characters is input. a histogram creation unit (34) that writes the distance data as an address and the number of occurrences of the distance data as data in a histogram area of the memory (18); and reads the memory in ascending order of addresses to create a histogram of each address. Cumulative output specified data count (OTD)
A threshold value determination circuit (36) that stops reading when the address becomes the threshold value, reads the distance data area of the memory, compares the read distance data with the threshold value, and selects the one that is smaller than the threshold value. A character recognition circuit comprising: a data output determination unit (38) that uses as an output candidate character.