JP2881318B2 - Conversion circuit of distance value in character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention performs a distance calculation based on a distance value conversion circuit, in particular, feature data of a recognized character and feature data of each character in a dictionary, The present invention relates to a distance value conversion circuit in a character recognition device in which distance data represented by N bits is represented by reduced n (n <N) bits.

[Related Art] In the process of performing character recognition, feature data x (i), (i = 1, 2,...) Obtained from a character pattern of a recognized character.
, And a characteristic data yj (i) given in advance for each character yj (j = 1, 2,...) In the dictionary, calculate a Euclidean distance Vj for each character. It is determined as a candidate character for the recognition character.

As an example of a calculation formula for calculating the Euclidean distance Vj, for example, Can be obtained by

In this case, as is apparent from equation (1), as the number i of the feature data to be compared increases, the value of the distance also increases, and it is necessary to display the distance data with N bits capable of displaying the maximum value. .

Conventionally, the distance Vj between the characteristic data x (i) of the recognized character and the characteristic data yj (i) of each character yj in the dictionary has been uniformly expressed by N bits.

[Problems to be solved by the invention]

Therefore, the memory for storing the distance expressed by N bits and its peripheral circuits must all be N bits, and there is a disadvantage that the circuit scale becomes large.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned drawbacks. In a character recognition device, the distance of feature data of a character that is a candidate character with respect to a recognized character is small, and the distance of feature data of a character that cannot be a candidate character is small. Note that a character having a distance greater than the maximum value represented by the predetermined number of bits cannot be a candidate character, so the distance for the character is determined by the predetermined number of bits. A character recognition device capable of expressing a distance value with a small number of bits by substituting the maximum number (when the distance is calculated as a positive value) or the minimum number (when the distance is calculated as a negative value). It is an object of the present invention to provide a distance value conversion circuit.

[Means for solving the problem]

In order to achieve the above object, the distance value conversion circuit in the character recognition device of the present invention includes a feature data for a recognized character and a dictionary data in the dictionary as shown in the basic configuration diagram of FIG. In a character recognition device that calculates a distance based on feature data of each character and obtains distance data expressed by N bits, the distance data expressed by the N bits is converted into a predetermined number K (= ^2α ). Lower n- of the number obtained by dividing by
A bit width reducer 1 for outputting n bits of 1 bit (n <N) and a sign bit, and a number obtained by dividing the distance data expressed by the N bits by a predetermined number K, the number of which is n bits or more An overflow / underflow detector 2 for detecting the presence or absence of valid distance data for a bit, a replacement data generator 3 for generating specific data with the same n bits as the output of the bit width reducer 1, an overflow / underflow When the detector 2 outputs a flow, the replacement data generator 3
And a selector 4 for outputting n-bit data of the bit width reducer 1 when no flow is output, and n-bit data obtained by reducing the distance data expressed by N bits. It is configured to express.

[Action]

Distance data N bits including a sign bit in the most significant bit is inputted to the bit width subsampling circuit 1 is divided by a predetermined number K = 2 ^alpha for reducing the bit width. That is, in FIG. 1 (II), the N-bit distance data is shifted by α. The bit width reducer 1 includes n-1 uppermost code bits, which are hatched, and n-1 bits from a least significant bit 0 to a bit n-2 obtained by dividing by the predetermined number K. The n bits are transmitted to the selector 4.

The replacement data generator 3 receives the most significant sign bit of the N-bit distance data, and when the sign bit indicates positive, it is represented by n-1 bits together with the sign bit. A maximum number (positive maximum value) is created, and when the sign bit indicates a negative value, a minimum number (negative minimum value) represented by n-1 bits is created together with a bit indicating a negative value. Input n bits.

When valid distance data (sign bits are not regarded as valid distance data) exists in n or more bits of a number obtained by dividing the N-bit distance data by a predetermined number K, the recognition character and the dictionary are used. The distance to the characters in the given number K
Means that the number obtained by dividing by n is too large to be represented by n-1 bits. A signal indicating a flow is output from the overflow / underflow detector 2 to the selector 4 and generated from the replacement data generator 3. The n-bit data described above is output from the selector 4. The n-bit data generated from the replacement data generator 3 is a number indicating that the distance is too large to be a candidate character of a recognized character, and indicates a flow from the overflow / underflow detector 2. Even if the distance data at which the signal is output is reduced to n bits, there is no effect.

When no flow signal is output from the overflow / underflow detector 2, the data reduced to n bits by the bit width reducer 1 is output from the selector 4. Since the data reduced to n bits of the bit width reducer 1 output from the selector 4 is uniformly divided by a predetermined number K, even if the data is reduced to n bits, the size between each data is small. The relationship is kept constant.

Therefore, N-bit distance data can be reduced to n bits,
The size of the circuit can be reduced.

〔Example〕

FIG. 2 shows the configuration of an embodiment of the present invention.
Corresponds to that of FIG. 1, 11 is a barrel shifter, 12 is a priority encoder, 13 is a comparator, 14 is an inverter,
15 is an ALU and 16 is an inverter.

Barrel shifter 11 corresponds to the bit width subsampling device 1 of FIG. 1, receives N-bit range data, i.e., the distance value, divided by the predetermined value K = 2 ^alpha to reduce a large value,
From the least significant bit 0 of the resulting number, n-1 bits of bit n-2 and bit n-
A total of n bits of 1 are output.
The most significant bit n-1 is the same as the most significant bit N-1 input to the barrel shifter 11.

The priority encoder 12 receives the N-bit distance data, searches the most significant bit N-1, that is, a value different from the value of the sign bit, from the most significant bit N-1 side.
This circuit outputs a number up to one bit position before the bit position of the different value, that is, a number m in which the same value as the most significant bit N-1 continues.

The ALU 15 is N- (n-1) obtained by subtracting the output bit number n-1 of the N-bit distance data inputted to the barrel shifter 11 excluding the sign bit from the N, and the predetermined number K = 2.
^{From the} power α of α through the inverter 14, N− (n−1)
This is an arithmetic circuit that calculates -α.

The comparator 13 compares m input from the priority encoder 12 with N− (n−1) −α input from the ALU 15, and m ≧ N− (n−1) −α (2) ), The selector 4 is selected by the barrel shifter 11 without outputting a flow signal. When m <N− (n−1) −α (3), a flow signal is output, and the selector 4 is replaced with replacement data. The operation is performed so as to cause the generator 3 to select.

The replacement data generator 3 outputs the N-bit distance data,
The most significant bit N-1 representing the sign bit is input, and n-1 numbers from bit 0 to bit n-2 inverted by the inverter 16 are created, and the most significant bit n
The number of the input most significant bits N is used as it is for -1, so that n-bit data is generated. That is, when the sign bit n-1 indicates a positive "0", the positive maximum value "111...
1 ”, and when the sign bit n−1 indicates a negative“ 1 ”, the negative minimum value“ 000 ”represented by the n−1 bits
... 0 "is generated.

 The operation of FIG. 2 will be described with reference to FIGS. 3 and 4.

If the distance data of N bits is input to the barrel shifter 11, the distance value of the N bits are divided by a predetermined number K = 2 ^alpha,
3 and 4, the α bit out of the N bits disappears and the (n−1) bit is replaced by the barrel shifter 11 in FIGS.
Are output to the selector 4 as bits 0 to n-2. At this time, as the most significant bit n-1 from the barrel shifter 11 to the selector 4, the same number as the most significant bit N-11 input to the barrel shifter 11 is transmitted as it is.

Further, as described in the replacement data generator 3, the most significant bit N of the N-bit distance data is stored in the selector 4.
A positive maximum value or a negative minimum value represented by n-1 bits corresponding to the number of -1 is input together with the sign bit n-1.

On the other hand, the priority encoder 12 of the overflow / underflow detection circuit 2 has a continuous number m of the same value as the bit N.
, And the comparator 13 calculates the number of continuations m detected by the priority encoder 12 and the N− calculated by the ALU 15.
(N-1) -α. Computed N- (n-1) -α This ALU15 from bit N-alpha obtained by dividing the distance data of N bits input to the barrel shifter 11 by a predetermined number K = 2 ^alpha to the selector 4 It represents the number of upper bits minus the lower n-1 bits being transmitted.

Then, m and N− (n−1) −α are compared by the comparator 13.
When the relationship shown in the equation (2) is satisfied, the effective distance value bits (n-1 bits) transmitted from the barrel shifter 11 to the selector 4 as shown in FIG. (Hatched). Conversely, it means that no effective distance value bit appears in the bit of N- (n-1) -α.

Therefore, in this case, it indicates that the distance between the recognized character and the character in the dictionary can be represented by n-1 bits, and the flow is not output from the comparator 13 but transmitted from the barrel shifter 11 to the selector 4. The n bits (bit n−1 is a sign bit) pass through the selector 4 as it is, and the n bits are output from the selector 4.

In addition, when the relationship is expressed by the equation (3), as shown in FIG. 4, the effective N- (n-1) -α bits which are not transmitted from the barrel shifter 11 to the selector 4 Bits (shaded) appear.

Therefore, in this case, it indicates that the distance between the recognized character and the character in the dictionary is farther than the maximum value represented by n-1 bits, the comparator 13 outputs a flow, and the selector 4 The n bits (bit n-1 is a sign bit) generated from the data generator 3 are output.

That is, in the case of Expression (3), since the distance from the recognition character is too large and cannot be a candidate character, the maximum value represented by n-1 bits (when bit n-1 is positive) or the minimum value This means that the replacement with a value (when bit n-1 is negative) has no effect on the selection of candidate characters.

The priority encoder 12 in the overflow / underflow detector 2 receives the N-bit distance data and searches the most significant bit N-1, that is, a value different from the value of the sign bit, from the most significant bit N-1. The number up to one bit before the bit position of the different value, that is, a number m in which the same value as the most significant bit N-1 is used, is used. The number of bit positions of the different values that are found may be m. In this case, the comparator 13 does not output a flow signal when m> N− (n−1) −α (4), and causes the selector 4 to be selected by the barrel shifter 11, so that m ≦ N− (n− 1) -α (5) A signal of a flow is issued to make the selector 4 select the replacement data generator 3 side.

In the former case, the priority encoder 12 sends the comparator
Since the number m output to 13 is smaller by 1 than the latter case, there may be a case where the number of bits may be small, and the scale of the circuit shown in FIG. 2 may be small.

As described above, the overflow / underflow detector 2 is the second
The present invention is not limited to the configuration shown in the figure, and when there is valid distance data in n bits or more of the number obtained by dividing the input N-bit distance data by a predetermined number K, the flow And any configuration may be used as long as the circuit switches the selector 4 to the replacement data generator 3 side.

The most significant bit n output from the barrel shifter 11
-1, the most significant bit n-1 of the replacement data generator 3
May be used.

〔The invention's effect〕

As described above, according to the present invention, N-bit distance data can be replaced with n (n <N) -bit distance data, the circuit scale can be reduced, and the recognition character collating process can be performed. Speed can be increased.

[Brief description of the drawings]

FIG. 1 (I) is a basic configuration diagram of the present invention, FIG. 1 (II) is a diagram for explaining bit reduction of a bit width reducer, FIG. 2 is a configuration of an embodiment of the present invention, FIG. 3 and FIG. FIG. 7 is a flow output explanatory diagram for explaining that a flow of an overflow / underflow detector is output. In the figure, 1 is a bit width reducer, 2 is an overflow / underflow detector, 3 is a replacement data generator, 4 is a selector,
11 is a barrel shifter, 12 is a priority encoder, 13
Is a comparator, 14 is an inverter, 15 is an ALU, and 16 is an inverter.

Claims (1)

(57) [Claims] 1. A method of reducing distance data represented by N bits obtained by distance calculation based on characteristic data of a character to be recognized and characteristic data of each character in a dictionary, n (n <n
A distance value conversion circuit in a character recognition device that is represented by N) bits, wherein the distance data represented by N bits is converted into a predetermined number K (=
2 ^α ), a bit width reducer (1) that outputs n bits of the lower n−1 bits and the sign bit of the number obtained by dividing the distance data expressed by the N bits by a predetermined number K An overflow / underflow detector (2) for detecting the presence or absence of valid distance data for n bits or more of the number obtained by dividing the number, and generating specific data with the same n bits as the output of the bit width reducer A replacement data generator (3) for causing the overflow / underflow detector to output n-bit data generated by the replacement data generator when a flow is output, and to reduce the bit width when no flow is output. And a selector (4) for outputting n-bit data of the device.