JPH03214284A - Data conversion system - Google Patents

Abstract

PURPOSE:To increase the processing speed in a line thinning operation, etc., by deciding the output data based on the value of the data on an area except the specific one and outputting the output data when an instruction signal is applied from a memory. CONSTITUTION:An arithmetic means 3 outputs its data when an instruction signal is outputted from a memory 2 so as to obtain a white or black center dot. Then the means 3 decides whether the center dot and its periperal ones are white or black with application of the signals which point the peripheral dots. The cells are formed in a systolic array and therefore the data D3 outputted from the means 3 is also applied to the corresponding cell of the next stage as well as to the cell contiguous to the corresponding cell, and the subsequent contiguous cells. Thus the line thinned data can be obtained from the cell as long as the line thinning data, for example, is stored in the memory 2.

Description

[Detailed Description of the Invention] [Summary] Regarding a data conversion method that performs line thinning of patterns for extracting features, the present invention speeds up the process of line thinning by processing in parallel in a pipeline using a system link array. The purpose is to capture data conversion methods, and multiple data are added.
a shift register that sequentially shifts the data; and data in a specific area of the shift register is added as an address;
When the data in the specific area has a pattern that allows the center value of the area to be determined, predetermined data is output, and when it cannot be determined, data is stored that outputs an instruction signal determined by data outside the specific area of the shift register. a memory for selecting and outputting data when no instruction signal is applied from said memory, and determining and outputting output data based on the value of data outside the specific area specified by the instruction signal when an instruction signal is added. and a calculation means for performing the calculation.

[Industrial application field]

The present invention relates to a pattern recognition device, and more particularly to a data conversion method for thinning a pattern for extracting features.

(Prior Art) With the development of computer systems, reading devices have been put into practical use that capture image data, cut out characters from the captured image data, and recognize each character in the text of a read document. This reading device divides the dot data read by an image scanner or the like into predetermined area units, and compares the characters within the division with the predetermined characters.
The most similar character is output as the result.

This predetermined data is generally stored in a dictionary memory, and is, for example, data characterizing each prescribed character. When the characters to be recognized are entered,
Similarly, the input character is characterized, and the distance from the predetermined characteristic data stored in the dictionary memory mentioned above is determined. The smallest character from this determined distance is output as the recognition result.

In the system described above, all of these processes are performed dot by dot. For example, characters within one square are cut out to be recognized characters, a 3 x 3 window is provided to obtain feature data from the characters, and features are obtained from the 3 x 3 dot data. This 3×3 window is read dot by dot by processing by a CPU or the like.

Alternatively, if the memory can be accessed in 8-bit units, three adjacent pieces of data (for example, 8 dots of data in one row) are read in one access, and then the next row's data is read... The reading is performed three times as shown in the following.

Furthermore, when all the feature data in the window is calculated,
From that feature data, the dictionary memory is sequentially accessed character by character, and the distance from each feature data is determined. Then, the distance from the feature data is compared character by character, and the character with the closest distance is output as the recognition result. Alternatively, the data of the top three or five characters is output as candidate characters.

[Problem to be solved by the invention]

There are various ways to obtain the above-mentioned feature data, but one method is, for example, when the input is data such as characters, eliminate the thickness of the characters and draw each character into one thin line (one continuous line of one dot). There is a method in which a character is represented by a character, and its features are extracted based on the vector direction of a single line, that is, a thin line, and the input character is recognized. This method is
The data manually generated in advance is normalized, and when the input data has the same size, the vector direction, that is, the feature vector, is compared with the previously stored character.

As mentioned above, this method requires a wire thinning circuit for thinning the wire.

Conventionally, this line thinning circuit determines whether the dots in the vertical and horizontal directions are black or white, centering on the center dot of, for example, a 3 x 3 dot within a character frame that has been cut out, and then determines whether the dots in the vertical and horizontal directions are black or white. It is determined whether or not to make the dots white.

To make this determination, multiple dots in the upper, lower, left, and right directions (total 8 directions) of the central dot must be read dot by dot, and it takes a lot of time to recall multiple dots to determine one dot. I had a problem.

That is, you have more time to thin the wire, so
Recognition processing such as character recognition takes a lot of time.

SUMMARY OF THE INVENTION An object of the present invention is to provide a data conversion method that speeds up processing such as line thinning by performing parallel processing in a pipeline using a system litho array.

[Means to solve the problem]

FIG. 1 is a block diagram of the principle of the present invention.

For example, a part of image data is added to the shift register l, and the data is sequentially shifted.

The memory 2 adds data of a specific area among the data shifted by the shift register 1 as an address, and when the pattern is such that the center value of the specific area can be determined, it outputs predetermined data, and when it cannot be determined, the data is added to the data shifted by the shift register 1. Data for outputting an instruction to be determined based on data other than the specific area within 1 is stored.

The operation stage 3 uses the memory to select and output data D1 when no instruction signal is applied, and when an instruction signal D2 is added, determines the output data based on the value of the data of the feature area specified by the instruction signal D2. and output D.

[Function] The shift register 1, memory 2, and calculation means 3 form one cell, and the cell constitutes a system litho array. The shift register 1 mutually adds adjacent data among the data added to each cell constituting the system litho array. That is, at least the shift register 1 receives, for example, three dot data in one dot row. Alternatively, adjacent data may also be added.

By shifting with the shift register l, for example, when one dot unit is added to the input of the systolic array of the cells, each cell sequentially shifts the added dots and adds data in the corresponding specific area to the memory 2.

The memory 2 stores in advance data for outputting the center dot corresponding to the data added to those dots, and for example, in the case of thinning, what kind of data are the eight dots surrounding the center dot? Set the center dot to 0 by
Or decide whether to set it to 1 (white or male). If the center dot is to be white or black based on the 8 pieces of data, but it cannot be determined, this is stored in the memory 2, and in this case, in addition to the above 8 dots, the outer cells are white. It outputs a signal instructing the dot to be judged based on whether it is black or black.

That is, the memory 2 has an output method that not only determines the center dot using the eight dots but also indicates other dots if the center dot cannot be determined using the eight dots.

When a signal indicating that the center dot is white or black is output from the memory, the calculation means 3 outputs the data, and when a signal indicating surrounding dots is added, the center dot including the surrounding dots is white. Decide whether it is black or white.

Since the cells are arranged in a system link array, the data D3 outputted from the calculation means 3 is further applied to the corresponding cell in the next stage, and also to the adjacent cell of the corresponding cell, and also to the adjacent cell.

Through the above-described operation, if data to be thinned, for example, is stored in the memory 2, thinned data can be obtained from the cell. Also, since it is configured as a system litho array, for example, thick parts such as letters can be
Thinning can be done in units of dots, and by providing a plurality of stages, it is possible to finally obtain a thinned character represented by a thin line consisting of one dot.

〔Example〕

The present invention will be explained in detail below using the drawings.

FIG. 2 is a system configuration diagram of an embodiment of the present invention.

Information read by an image scanner or the like is stored in the image memory 10 as image data.

This image memory 10 has a storage capacity for one page read by an image scanner, and stores each dot of the read information as binary data of white or black, that is, 0 and 1 data.

The image data stored in the image memory IO is applied to a noise removal module 11 to remove noise generated during reading. For example, the noise removed by this noise removal module 11 is noise unrelated to character information, etc. For example, a 3 x 3 mask with a black center and 8 pixels surrounding the center dot.
The dots are noise such as white, and the noise removal module 11 makes the dot in the center white. Although this noise removal module is provided in the character recognition pre-processing section 12, it is not limited thereto, and may be performed when storing each character in the normalization module 16, which will be described later. It may be performed at the time of line element formation.

The image information from which noise has been removed by the noise removal module 11 is sent to the row histogram module 13, the column histogram module 14, and further to the readout control module 1.
Join 5. The row histogram module 13 is a module that projects the read information, for example, the content of the paper read by the above-mentioned image scanner, in the column direction in units of dots, and calculates the number of rows in each column.

That is, this is a process of determining how many black dots are present in one dot row (horizontal direction) for each dot row. The column histogram 14 is a process of projecting in the column direction and calculating the number of projected black dots in the same way as the row histogram described above.

Data is sequentially read out dot by dot in the row direction from the image memory 10 and added via the noise removal module 11 (reading of dots similar to raster scanning), and the row histogram module 13 sequentially counts black dots (1 dot line). Then, the number of black dots is sequentially calculated for each row. This number of black dots becomes the row histogram corresponding to each row. The column histogram 14 also has counters corresponding to the number of dots in one dot row.
Each time a row of dots is added in sequence, the counter corresponding to the black dot is incremented. By performing the above-described operations for one page, the row histogram module 16 and the column histogram module 14 obtain so-called row histograms and column histograms representing the number of dots for row positions and column positions, respectively. The result is then applied to the read control module 15.

The readout control module 15 sequentially obtains row positions and column positions from these row histograms and column histograms. For example, this position can be obtained by the period of the row histogram or the period of the column histogram.

The read control module 15 determines the row and column positions, but also performs the following processing. Image data, for example, information read from an image scanner, may have a tilt depending on the position of the paper and the like.

For this reason, the readout control module 15 sequentially changes the angle at which the histogram is obtained so that the column histogram and the row histogram take the maximum value, and obtains a correction angle.

Then, input the image information added from the above-mentioned noise removal module 11 again to obtain the final histogram,
From the point where the row histogram obtained by the corrected slope (the histogram takes the maximum value) changes from 0 to positive (possibly from positive to 0), one line of data corresponding to the slope is read out for one period. The data is stored in a row huffer provided in the control module 15.

The read control module 15 further obtains the column histogram within the row of one row of data stored in the row bath sofa, cuts out the data from the position where the column histogram changes from O to positive, and outputs it to the normalization module 16. It is also output to the conversion table creation module 17.

This extracted data is data for a single character area.

The conversion table creation module 17 is a module that obtains conversion data for normalizing one character by the normalization module 16. It projects the conversion data in the column direction and the row direction on the one character area cut out by the readout control module 15, and Counters in the column and row directions are incremented dot by dot (row and column) starting from the column and row where the dot exists, and the value up to the final value within the area of one character is determined.

In the normalization module 16, the final value of the dot cut out by this one character in the row direction and column direction and the cut out 1
Based on the size of the character, the character is expanded to cover the entire area within the extraction area. For example, an enlargement process is performed to make an area of 64×64 dots into one character area. If the value in the column direction and row direction of the character is 48 dots (both column and row) in the conversion table creation module 17,
Performs processing to convert 48 dot characters to 64 dots.

In this process, data in a row or column at a specific position is repeated to make the same data and enlarge the characters. Furthermore, in the case of reduction, rows and columns at specific positions are repeatedly read out and ORed together to reduce them as the same row or example.

The normalization module 16 allows one character area, for example 64X
After one character is enlarged within 64 dots, the thinning module 18 performs a process of thinning the character. In this thinning module 18, one dot above, below, left and right of the center dot is
(3X3) and 1 dot to the left of it and 2 dots above from the center
Line thinning processing is performed using a mask with a total of 11 dots. Moreover, this mask can also be performed using 9 dots of 3×3.

By controlling the central dot to be O when the pattern is predetermined by the mask described above, thinning of lines around one dot forming a character can be achieved in one process. By sequentially repeating this thinning of the mask, a character can be formed by a one-dot line.

For example, 64×6 obtained by the thinning module 18
The 4-dot thin line character is added to the line element generation module 19 and converted into line elements. In this line elementization module, when black dots exist in the vertical direction from the target dot, that is, the center dot, when they exist in the horizontal direction, when they exist in the upper right and lower left, and when they exist in the upper left and lower right, Each dot is represented by a total of four types of line elements. In addition, the above 4
If the line element belongs to more than one of the types, priority is given in the order of, for example, the vertical direction, then the horizontal direction, etc., and in which direction the line element exists is determined for each dot unit. Note that if the center is 0 dot, that is, white, it is assumed that no line exists.

In the line segmentation module 19, there are four directions: up and down, left and right, diagonal upwards to the right, diagonally upwards to the left, and five types, including the case where there is no line element, so the state is expressed in a 3-bit value for each dot. , a total of 3X64X64 information,
Add to feature vector module 20.

In the feature vector module 20, the line segmentation information obtained in the line segmentation module E9 described above is divided into units of 8 dots in the left, right, top, and bottom, and the divided areas are divided into one area downward and one area to the right (2 × 2 areas), a total of 16 dots is defined as one vector module area, and how many vertical, horizontal, horizontal,
It is counted whether there are line elements in four directions: upper right direction and upper left direction. The feature hector is calculated using an area of 16×16 dots as one vector module area. Since this one vector module area is moved in units of 8 dots, there are 7 areas in each of the row and column directions, and a total of 7×
This is the area of 7 feature vectors.

In the feature vectorization module 20, the number of directions is calculated for each area as described above, but when calculating this number, weighting is applied to each area, with the center being higher and the surrounding areas being lower as they move outward. There is. For example, each dot in the center 4x4 area has a weight of 4, each dot in the 2 dots around it has a weight of 3, each dot in the 2 dots around it has a weight of 2, and then the 2 dots around it have a weight of 4. Each dot is set to 1, weighting is performed, and a feature vector is determined.

This feature vector has specific characters to be recognized all made the same size by the normalization module 16, so
If they are the same characters, they have almost the same characteristic hector,
The feature vectors differ for each character. However, since there are very similar modules, in the embodiment of the present invention, in order to speed up the calculation process and improve the recognition rate, a standard pattern of feature hectors is used for each feature vectorization region, i.e. Classification is performed within each square, and the distance between the 20 class standard patterns and the added unknown input within each square is determined. That is, the distance between the feature vector in each square of the standard pattern and the feature vector in the square obtained by the special notification vector module 20 is determined for each square. Each of the squares is divided into classes (classes 1 to 20), and the distance ranking of the classes within each square is determined in descending order of distance to the fifth class.

The distance calculation module 21 stores this distance in the class dictionary 23-
1 (standard patterns are stored in class units). Note that when searching for individual candidate characters, the candidate dictionary 23-2 is used (at this time, the switch SW selects the candidate dictionary 232).

The top selection and score allocation module 22 determines the top five classes mentioned above and determines the score corresponding to each class for each square. In other words, the top selection & score assignment module 22 determines the score to be given to each of the 5th to 5th classes in class based on the distance obtained from the distance calculation module 21, and calculates the score of each character. For example, for the first distance (short distance), 5 points are given, then 4 points, and 3.2.1 points are given to the class. These are provided corresponding to squares 1 to 49, respectively. The processing results of the top selection score module 22 are added to the comprehensive evaluation module 24.

The comprehensive evaluation module 24 is a module that calculates the degree of matching between an input object, that is, an input character and its candidate, and has three types of operation: an associative matching mode, an exhaustive matching mode, and an individual matching mode.

The associative matching mode is a mode in which the score of a candidate is calculated from the mask corresponding to the candidate stored in the associative dictionary 23-3 and the class to which the candidate belongs.

As shown in Figure 2(b), the associative dictionary has a candidate ID for each mask.
The class ID of the class to which the candidate belongs in the mask is stored, using the address as the address.

This data is the Cdim corresponding to each candidate's mask ID.
It is obtained by clustering a set of dimensional subvectors according to their (weighted) distances, and only the results are stored in an associative dictionary.

Class dictionary 23-1 in the distance calculation module at the same time
is also created correspondingly.

Note that the associative dictionary 23-3 and the class dictionary 23-1 correspond to each other and have the same type. When two or more types of dictionaries are stored in one memo, the specification of the dictionary to be used becomes the dictionary reference start position. (This dictionary can be divided into candidate IDs and a comprehensive evaluation can be performed on each in parallel.
If higher speed is required, it can be easily realized).

The associative dictionary 23-3 is a table in which the class ID: K of the class to which candidate a belongs with mask m is written, and this is written as C (m,
When expressed as a)=K, for candidate a(-1 to c cand), it is obtained as follows. Note that P (m, k) represents the score here. Based on this formula, the overall evaluation value V (a
).

The total matching mode and individual matching mode of the comprehensive evaluation module are modes in which calculations are made for each candidate.

The total matching mode is a=1-c cand, the individual matching mode is J-1~c kind, a=b(j),
Express the distance by d(m, a). This value V (a) is the (weighted) distance of the feature vector between candidate a and the input object.

The top candidate selection module 25 selects and outputs a plurality of characters, for example, five characters determined from the top in each character correspondence.

The top five characters become the recognition result in the read image data.

All of the operations described above are performed by pipeline processing. That is, data for one page, for example, in the image memory 10 that stores image data is read out by pipeline processing, divided into lines by the control module 15, and outputted to the normalization module 16 in units of characters. The top selection module 25, which performs the above-mentioned thinning, line elementization, feature vectorization, and recognition processing on the single character, is a module that ranks candidates based on the comprehensive evaluation value and selects the top five.
If the input is in associative exhaustive matching mode, then I (a',
V(a) l a', a = 1 ~c cand
} If it is the discrete integer sum mode ( (j. v (a) lj : 1 ~ c kind
, a = b (j)) (Comprehensive evaluation output of individual matching) Descending/ascending order: (Character association: Largest order, Others: Smallest order). Further, the output is the candidate IDs (or input order) arranged in the order of the input sort results and their comprehensive evaluation values.

In the embodiment of the present invention using FIG. 2 described above, a character recognition device including a line thinning module was explained.

The thinning module will be further explained in detail below.

FIG. 3 is a configuration diagram of a thinning array according to an embodiment of the present invention. The pattern data within one frame that has been normalized by the normalized module 16 is processed by the thinning module 1 in line units.
Add to 8. The thinning module 18 is composed of thinning cells H(11) to H(L,N) as shown in FIG. Each thin line cell H (L, N) adds 1-binoto and 2-binoto i-data in the vertical direction, that is, to adjacent cells. FIG. 4 is a detailed configuration diagram of the thinning cell.

Dot data D added to the thinning cell (L, N).

is added to shift register RSI. shift register RS
I are registers RIO, R2, R8 . This is a 4-stage R6 shift register. Moreover, D0 is also added to the adjacent cell. On the other hand, data added from the adjacent cells is added to shift registers RS2 and RS3.

Data from above in FIG. 4 is applied to shift register RS2, and data from below is applied to shift register RS3. Shift register RS2 is shift register R3, R
4, R5, shift register RS3 is shift register Rl
, RO, and R7. Data added from cells further below is added to shift register R9. The output of shift register R7 is also applied to shift register RO, but the result is also output to the upper cell in FIG. 4. That is, the data added to the shift register R9 is added from the data shifted by one stage in the lower cell in FIG. If we summarize the data added from the horizontal direction, the data input to each cell will be dots in 8 directions centering on R8, and 1 dot further to the left, as shown in the explanatory diagram of the register and pattern position as shown in Figure 5. It is composed of a total of 11 dots, including I dots further down.

Shift registers R6, R8, R2, R5. R4, R3,
R7, RO. The nine outputs of Rl are applied to the address input terminals of memory F. Memory F has a 3-bit output, which is applied to buffer B.

Among the data DIO, Dll, and DI2 applied to the Huffer B, the data DIO is applied to the input of the OR gate OR1.

As shown in FIG. 5, when it is determined that the center dot (stored in shift register R8) should be white or black using 8 dots surrounding the center dot (stored in shift register R8), data DllD12 becomes O and DIO It becomes the value of the decision.

That is, if it is black, it is set to 1 and is output via the OR gate ORl. If it is white, it is output as O via the OR gate OR1.

On the other hand, when the value of the center dot l-R8 cannot be determined from these 8 dots, the memory F sets the data Dll or D12 to 1.

Data Dll and D12 are AND gate ANDIAND2
Participating in As described above, if the center dots 1-R8 have been determined, the data D11 and D12 are O, but if they cannot be determined, the values of other dots are taken into consideration to obtain the result.

Data Dll is an instruction signal that takes into consideration the value of register R10 and indicates the result, and data D12 is an instruction signal that determines the data based on the value of the dot of register R9.

The output of register RIO is applied to ANDI via inverter INVI. Also, register R
The output of 9 is connected to AND gate A via inharter INV2.
．． Joining ND2. That is, data Dll
When is 1, the dot data stored in the register RIO is inverted and the output is determined based on the result. Also, data D
When 12 is 1, it is determined by the value of the output of register R9.

In other words, if register R9 is 0, Dll
When is 1, 1 and 1 are added to the two inputs of the AND gate AND1, and it outputs ■.

Further, when the AND gate D12 is 1 and the value stored in the register R9 is O, 1 and 1 are added to the AND gate AND2, and the AND gate AND2 outputs 1, that is, the H level.

The output of AND gates ANDI and AND2 is OR gate O
R2, and when one of their results is 1, the OR gate OR2 outputs l. OR GATE OR
The output of D2 is input to the OR gate ORl, and as a result, the output D20 outputs the logical result of the AND gates ANDI and AND2.

In other words, the data in register R8 is set as the center dot, and the surrounding data (registers RO, Rl,
R2, R3, R4, R5, R6, R7), register R9 or register RI
It is determined by the value of O.

FIG. 6 is an input data diagram, and FIG. 7 is an explanatory diagram of data changes in the input data register.

At time L1, when a is added to cell J+1, b is added to cell J, C is added to cell J-1, and d is added to cell J-2, b is added to register RS1 of cell J, and a is added to register RS2 of register RS3. C is added to.

Then, C is stored in register RSI of cell J-1, b is stored in register RS2, and d is stored in register RS3. Also, at time t2, e, f, g, and h are added, so the shift register is sequentially shifted, b and f are added to register RSI, a and e are added to register RS2, and register RS3
c and g are added to register 9, and d is added to register 9, and each register shifts and stores its value. In addition, in cell J-1, c and g are in register RSI, bf is in register RS2,
d and h are added to register RS3 and their values are stored.

After shifting sequentially as described above, i, j.
k,l are further mn.k and l at time t4. When o and p are added, cell J at time t4:yo register RSl?
．． :b, f, j, n, register R S 2 6: e
, i. m, g. in register RS3. k, O, Le]
1 is stored in 2 star R9. Also, c, g, k, o are in register RSI of cell J-1, and f is in register RS2.
, j, h, and h, z, p are stored in register RS3. That is, at time t4, data abcd, efgh,
When ijkj2, mnop are added, the data is transferred to registers RSI, RS2 . It is stored in RS3 and RE9. At this time, the center dot of the target cell is J.

As mentioned above, when the center dot cannot be determined based on the pattern of the center dot within a simple 3 x 3 dot, the calculation unit calculates the surrounding dots so that the desired line thinning can be achieved accurately. can.

〔Effect of the invention〕

As described above, according to the present invention, line thinning is performed sequentially by vibline processing and further in parallel, so that line thinning can be performed at high speed. Additionally: Pattern recognition can be faster.

[Brief explanation of drawings]

Fig. 1 is a block diagram of the principle of the present invention, Fig. 2 is a system configuration diagram of an embodiment of the present invention, Fig. 3 is a block diagram of a thinning array in an embodiment of the present invention, and Fig. 4 is a diagram of a thinning cell. FIG. 5 is an explanatory diagram of registers and pattern positions, FIG. 6 is an input data diagram, and FIG. 7 is an explanatory diagram of register data changes. DESCRIPTION OF SYMBOLS 1...Shift register, 2...Memory, 3...Arithmetic means.

Claims (1)

[Claims] 1) A shift register (1) to which a plurality of data is added and which sequentially shifts the data, data in a specific area of the shift register is added as an address, and the data in the specific area is located at the center of the area. When the pattern is such that the value of D can be determined, predetermined data (D
_1), and when it cannot be determined, outputs an instruction signal (D) determined by data outside the specific area of the shift register.
a memory (2) for storing data to output _2); and a memory (2) that selects and outputs data (D_1) when no instruction signal is added from the memory (2), and selects and outputs data (D_1) when an instruction signal (D_2) is added. A data conversion method characterized by having a cell comprising an arithmetic means (3) that determines output data based on the value of data outside the specific area indicated by (D_2) and outputs it (D_3). 2) the cell is connected to a systolic array structure, the cell applies a first data to an adjacent cell, and the plurality of data applied to the cell includes at least the first data applied to the adjacent cell; 2. The data conversion system according to claim 1, further comprising: said second data being added directly. 3) The plurality of data is dot image data representing black and white, the shift register (1) adds data of a specific area of 3 x 3 dots to the memory (2), and the memory (2) is a 3 x 3 dot image data. When the data of the center dot cannot be determined as white or black, the calculation means (3
), the calculation means (3) turns on the AND gate by an instruction signal corresponding to the instruction dot, and outputs data of the dot added to the turned-on AND gate as decision data. data conversion method.