JPH0420228B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to a method for extracting features of character/graphic patterns at high speed and with high accuracy.

(Prior Art) Conventional character recognition devices often employ a method of extracting strokes from a character pattern and recognizing them using the positions, lengths, mutual relationships between strokes, etc. of the extracted strokes. This type of device (1) calculates the curvature of a contour point series detected by tracing the contour of a character figure, divides the contour series using points with a large value of curvature as dividing points; (2) Extract the strokes by combining the series, or (2) perform thinning processing on the character/figure pattern to create a skeleton, and track the connectivity and skeleton pattern of the skeleton pattern to detect points of sudden angle change, etc. The strokes are extracted by detecting the strokes, and the geometric features and the like are extracted from the extracted strokes to identify the characters and figures. However, in method (1), when the character/graphic pattern becomes large or complex, the amount of processing increases, resulting in a decrease in processing speed. Furthermore, there are problems such as pattern distortion and whiskers due to thinning, and subsequent processing becomes complicated. In addition, in order to eliminate these defects, a method has been proposed in which scanning lines are drawn in each direction from each point in a character figure pattern, and the number of intersections between the scanning line and the character line is extracted as a feature of the point of interest. However, the feature of the number of intersections to be extracted varies depending on the slope of the character line, so the extracted feature is unstable, and the number of intersections alone is insufficient to reflect the structure of the character.

(Objects and Structure of the Invention) The present invention aims to improve these drawbacks by extracting a sub-pattern representing a stroke component in a desired direction of a character/figure pattern, and extracting a sub-pattern representing a stroke component in a desired direction from each point or arbitrary point of the sub-pattern. As a feature representing the distribution of character lines when looking at It is an object of the present invention to provide a feature extraction method that can reduce the degree of variation in the features in the prior art and can be used in a high-precision character recognition device by extracting these features and treating them as features of a character/figure pattern.

(Embodiment of the invention) FIG. 1 shows a character recognition device that is an embodiment of the invention. In FIG. 1, 1 is a photoelectric conversion unit, 2 is a pattern register, 3 is a line width calculation unit,
4 is a character frame detection unit, 5 is a division point determination unit, 6 is a vertical sub-pattern (hereinafter also referred to as VSP) extraction unit, 7
is a vertical sub-pattern memory, 8 is a VSP horizontal feature extraction unit, 9 is a VSP horizontal feature pattern memory, 1
0 is a horizontal sub-pattern (hereinafter also referred to as HSP) extractor, 11 is a horizontal sub-pattern memory, and 12 is a
HSP vertical feature extraction unit, 13 is HSP vertical feature pattern memory, 14 is right diagonal sub pattern (hereinafter referred to as
15 is a right diagonal sub-pattern memory; 16 is an RSP left diagonal feature extraction unit; 1
7 is an RSP left diagonal feature pattern memory; 18 is a left diagonal sub-pattern (hereinafter also referred to as LSP) extraction unit;
19 is a left diagonal sub-pattern memory, 20 is an LSP right diagonal feature extraction unit, 21 is an LSP right diagonal feature pattern memory, 22 is a VSP horizontal feature vector extraction unit,
23 is an HSP vertical feature vector extraction unit; 24 is an HSP vertical feature vector extraction unit;
25 is an LSP right diagonal feature vector extraction unit, 26 is an identification unit, 27 is a dictionary, and 28 is a character name output.

The operation of the first embodiment will be explained below.

First, the characters and characters on the form set in the reading mechanism.
The graphic pattern is converted into a binary quantized digital electrical signal in the photoelectric conversion section 1 and stored in the pattern register 2. At the same time, the line width of the input pattern is calculated in the line width calculating section 3. The vertical sub-pattern extractor 6 performs horizontal scanning over the entire pattern register 2, and calculates the continuous length and line width of black bits (character lines are black bits and character background parts are white bits). A vertical sub-pattern (VSP) is extracted based on the relationship with the determined line width and stored in the vertical sub-pattern memory 7. Similarly, the horizontal sub-pattern extraction unit 10, right diagonal sub-pattern extraction unit 14, and left diagonal sub-pattern extraction unit 18 generate a horizontal sub-pattern (HSP) by horizontal scanning, a right-diagonal sub-pattern (RSP) by scanning 45 degrees to the right, respectively. A left diagonal subpattern (LSP) is extracted by scanning diagonally left at 45° and stored in each subpattern memory. Figure 2 shows an example of the original pattern and each sub-pattern; (a) is the original pattern, (b) is the vertical sub-pattern (VSP), (c) is the horizontal sub-pattern (HSP), and (d) is diagonal to the right. subpattern (RSP), (e) is a left diagonal subpattern (LSP).

In this embodiment, the pattern register 2 and each of the sub-pattern memories 7, 11, 15, and 19 are 128 x 128-bit memories, that is, pattern memories with a width of 128 bits in the X direction and a width of 128 bits in the Y direction. be. The character frame detection unit 4 detects a rectangular frame (hereinafter referred to as a character frame) circumscribing the character figure pattern in the pattern register 2, and detects a rectangular frame (hereinafter referred to as a character frame) that defines the character frame in a two-dimensional plane defined by the pattern register 2. The position coordinates are sent to the division point determining section 5.

In the following description, a coordinate system with the origin at the lower left of a two-dimensional plane defined by the pattern register 2 will be used, with the horizontal direction being the X direction and the vertical direction being the Y direction. Further, in each sub-pattern memory, character line portions are designated as black dots, and other portions are designated as white dots.
The VSP horizontal feature extractor 8 extracts horizontal features for each point in the vertical sub-pattern memory 7 by performing horizontal scanning over the entire vertical sub-pattern memory 7.

FIG. 3 shows the configuration of a feature extraction unit such as the VSP horizontal feature extraction unit, and the connection relationship between the feature extraction unit, sub-pattern memory, and feature pattern memory. The area within the dotted line frame in FIG. 3 shows the feature extraction section, and here it is assumed that the VSP horizontal feature extraction section 8 is shown. Note that the HSP vertical feature extraction section 12, RSP left diagonal feature extraction section, and LSP right diagonal feature extraction section 20 have similar configurations. Figure 3
In the VSP horizontal feature extraction unit, 31 is a control circuit that controls the entire feature extraction unit, and 32 is a control circuit that refers to the contents of the sub-pattern memory while scanning the sub-pattern memory to detect a change point where the color changes from white to black. A changing point detection circuit, 33 a changing point memory for storing the X coordinate of a black point when it changes from a white point to a black point as a changing point, 34 a changing point counter for specifying the address of the changing point memory, and 35 a pattern memory. An address counter, 36 a comparator, and 37 an arithmetic circuit. Further, 38 and 39 are registers A and B, respectively, which are used when extracting features in two directions in each feature extraction section, for example, features in both left and right directions in the VSP horizontal feature extraction section. We will call them the direction register (REGL) and right direction register (REGR).

Using the flowcharts in Figures 3 and 4
The horizontal feature extraction operation in the VSP horizontal feature extraction unit will be described next. Figure 4 is a flowchart showing the process of calculating horizontal features regarding one point of interest, and CM(i) represents the contents of the changing point memory (to be described later) (i = 1, ..., I, I is the number of change points detected and registered in the change point memory in the first horizontal scan including the relevant point), x is the X coordinate of the relevant point, REGL is the left direction register, and REGR is the right direction register. It is.

When horizontal scanning is started from a point on the Y axis of the vertical sub-pattern memory 7 and a change from a white point to a black point is detected using the change point detection circuit 33, the black point (black and white change) when the white point changes to a black point is detected. (referred to as a point)
While registering the coordinates in the changing point memory 33,
Scanning is performed to the right end of the vertical sub-pattern memory 7 while incrementing the change point counter 34.

Next, start the horizontal scan again from the point where the horizontal scan started, and perform the fourth scan on all the points on the scanning line.
The process shown in the flowchart in the figure is performed. That is, first, the number of change points () detected in the horizontal scanning is set in the change counter 34 (step
401), 16-bit left register (REGL) 38
After initializing the right direction register (REGR) 39 (step 402), the X coordinate (fourth
X of the changing point which is the contents of the changing point memory using the contents registered in the changing point memory (indicated by CM(i) in the flowchart of FIG. 4) The comparator 36 determines the magnitude relationship between the coordinates and the X coordinate of the point of interest (the X coordinate output from the pattern memory address counter 35 used when scanning the vertical sub-pattern memory) (step 404), and the point of interest is determined by the comparator 36 (step 404). If the X coordinate of the point of interest is larger than the contents of the change point memory 33, that is, if the contents stored in the change point memory 33 indicate the position of the character line to the left of the point of interest, the X coordinate of the point of interest and the point of change The value of the difference with the content of the memory 33 raised to the K power (K is a constant; in this embodiment, K=2), that is, the distance between the point of interest and the character line raised to the K power, is calculated in the 16-bit left direction register (REGL) 38. Vessel 37
(step 405), and if the X coordinate of the point of interest is smaller than the contents of the change point memory, the difference between the X coordinate of the point of interest and the contents of the change point memory to the K power, that is, the distance K The process of adding the multiplication value to the 16-bit right direction register (REGR) 39 (step 406) is performed as shown in steps 403 and 407. The contents of the left direction register 38 and right direction register 39 extracted by incrementing the change point counter 34 that specifies the address of the point memory are used as the left direction horizontal feature and the right direction regarding the point of interest. As a horizontal feature, it is stored at a position in the VSP horizontal feature pattern memory 9 that corresponds to the position in the vertical sub-pattern memory 7 of the point of interest. FIGS. 7A and 7B are explanatory diagrams of VSP horizontal features, in which black circles indicate points of interest and white circles indicate points of black and white change. In FIG. 7A, the value of the left horizontal feature described above is for the case where a character line V ₁ exists at a distance of 10 to the left of the point of interest, and a character line V ₂ exists at a distance of 5 to the right of the point of interest. (100) and the value of the rightward horizontal feature (25),
FIG. 7B shows the case where the character line V ₃ exists on the left side of the point of interest, and the character lines V ₄ and V ₅ exist on the right side.

If we pay attention to the rightward horizontal feature, we can see that the feature value increases from 25 to 125 as the number of character lines increases.

In this way, the VSP horizontal feature extraction unit 8 performs horizontal scanning twice from each point on the Y axis on the vertical sub-pattern memory, extracts horizontal features for all points on the scanning line, and stores the horizontal features in the VSP horizontal feature pattern memory 9. Store. This VSP horizontal feature pattern memory 9 and each other feature pattern memory 13, 1
As shown in Fig. 5, the structure of 7 and 21 is a pattern memory with a width of 128 bits in the X direction and a width of 128 bits in the Y direction, and is specified by the X and Y coordinates of each feature pattern memory. Each point represents a feature (for example, left horizontal feature + right horizontal feature) and has a width of 32 bits. Figure 6 shows the characteristics of each point represented by these 32 bits. For example, it represents the 32-bit horizontal feature extracted for each point in the vertical subpattern memory, the upper 16 bits are the left horizontal feature, and the lower 16 bits represent the rightward horizontal feature. Also, HSP vertical feature extraction unit 1
2. The RSP left diagonal feature extraction section 16 and the LSP right diagonal feature extraction section 20 also perform the same operation as the VSP horizontal feature extraction section 8.

The outline will be explained below. HSP vertical feature extraction part 1
2 is a horizontal sub-pattern memory 11 in which the horizontal sub-pattern (HSP) extracted by the horizontal sub-pattern extraction unit 10 is stored, and in the process of performing vertical scanning twice from each point on the X axis, each point on the scanning line is The K sum of the distances between the character line that exists above the point of interest and the point of interest is the upward vertical feature, and the distance between the character line that exists below the point of interest and the point of interest. The sum of K powers is extracted as a downward vertical feature and stored in the HSP vertical feature pattern memory 13 having the same structure as the VSP horizontal feature pattern memory 9. The RSP left diagonal feature extraction unit 16 extracts each point on the Y axis and the right diagonal subpattern with respect to the right diagonal subpattern memory 15 in which the right diagonal subpattern (RSP) extracted by the right diagonal subpattern extraction unit 14 is stored. In the process of performing two 45° left diagonal scans from each point on the upper side of the memory 15, the distance K between the character line existing 45° to the left above the point of interest and the point of interest for each point on the scanning line. The sum of the powers is extracted as an upward left diagonal feature, and the K sum of the distances between the character line existing 45 degrees below the left of the point of interest and the point of interest is extracted as a downward left diagonal feature, and the VSP horizontal feature pattern memory 9 It is stored in the RSP left diagonal feature pattern memory 17 having the same structure as . The LSP right diagonal feature extraction unit 20 extracts the left diagonal subpattern (LSP) extracted by the left diagonal subpattern extraction unit 18 from each point on the Y axis and the X axis. In the process of scanning 45 degrees diagonally to the right from each point twice, each point on the scanning line is
The K sum of the distances between the character line that is 45 degrees above the right of the point of interest and the point of interest is the upward right diagonal feature, and the character line that is 45 degrees below the point of interest to the right and the point of interest are The K-th power sum of distances is extracted as a downward diagonal feature and VSP horizontal feature pattern memory 9
LSP right diagonal feature pattern memory 21 with the same structure as
Store in.

Although K=2 is used in this embodiment, good results can also be obtained with K=1 for characters with a relatively small number of character lines (also referred to as character dose), such as kana characters.

The dividing point determining unit 5 refers to the coordinates regarding the position of the character frame received from the character frame detecting unit 4, calculates the dividing point coordinates for dividing the area within the character frame into M×N partial areas, and calculates the dividing point coordinates for dividing the area within the character frame into M×N partial areas. feature vector extraction unit 22,
The calculated division point coordinates are sent to the HSP vertical feature vector extraction section 23, the RSP left diagonal feature vector extraction section 24, and the LSP right diagonal feature vector extraction section 25. The VSP horizontal feature vector extraction unit 22 uses the character frame position coordinates received from the character frame detection unit 4 and the division point coordinates received from the division determination unit 5, and uses the VSP
The area within the character frame in the horizontal feature pattern memory is M×
It is divided into N partial regions, and an M×N×2-dimensional horizontal feature vector whose elements are the arithmetic averages of the rightward horizontal features and leftward horizontal features of each point in each divided region is extracted. Hl _n,o HR _n,o (m= 1,...,M,n=
1,...,N), and the rightward horizontal feature and leftward horizontal feature of each point in the region are respectively expressed as l _x,y ,
When expressed as r _x,y and the total number of points in the area as T _n,o , the HL _n,o and HR _n,o are defined by equations (1) and (2).

HL _n,o =ΣΣl _x,y /T _n,o (1) HR _n,o =ΣΣr _x,y /T _n,o (2) HSP vertical feature vector extraction unit 23, RSP left diagonal feature vector extraction unit 24 , the LSP right diagonal feature vector extraction unit 25 refers to the HSP vertical feature pattern memory 13, the RSP left diagonal feature pattern memory 17, and the LSP right diagonal feature pattern memory 21, respectively, and extracts the VSP horizontal feature vector extraction unit 2.
Similarly to 2, the area within the character frame in each feature pattern memory is divided into M×N partial areas, and each of the vertical features in both directions, the left diagonal feature, or the right diagonal feature of each point in each partial area is divided. Calculating the arithmetic mean
In the same manner as the VSP horizontal feature vector extraction unit 32, M×N
×Extract a two-dimensional feature vector.

The identification unit 26 identifies the M×N×2×4-dimensional feature vectors extracted by the four types of feature vector extraction units and the feature vectors that are written in the same format as the feature vectors prepared in advance and stored in the dictionary 27. Calculate the distance between multiple standard character patterns, that is, the Euclidean distance between two vectors,
The character name of the standard character pattern whose distance gives the minimum value is output to the character name output 28.

As explained in detail above, the features to be extracted in the feature extraction method of the present invention are the distribution of character lines in each direction viewed from each point, that is, the extent to which the perpendicular line segment is to the right and left when viewed from each point. , and at what distance do horizontal line segments exist above and below, and at what distance do diagonal left and right line segments exist in the upper right, lower left, upper left, and lower right directions? Since the character dose in each direction is also reflected by the K-th power sum of the distances, it sufficiently reflects the geometric structure of the characters. In addition, in the process of detecting character lines in each direction, sub-patterns are used to focus only on character lines that are perpendicular to the relevant direction, so the influence of the slope of character lines due to differences in handwritten characters is ignored. stable feature extraction is possible. Furthermore, since the feature extraction for each point is performed only by a simple scanning process, there is an advantage that the apparatus can be made smaller.

(Effects of the Invention) The present invention uses subpatterns to extract the sum of K powers of the distances between a point of interest and a character line as a feature indicating the distribution of character lines in each direction as seen from each point of a character pattern. Therefore, it is a feature that reflects not only the distance distribution of character lines but also the character dose, and it is possible to perform stable feature extraction that fully reflects the geometric structure of the entire character. Since it is realized by a simple scanning process, it can be used in a high-speed and highly accurate character recognition device.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment of the character recognition device of the present invention, FIG. 2 is a diagram showing an example of an original pattern and a sub-pattern, and FIG. 3 is a block diagram of each feature extraction unit.
Figure 4 is a flowchart of horizontal feature calculation for each point, Figure 5 is a configuration diagram of each feature pattern memory, Figure 6 is a feature configuration diagram of each point in each feature pattern memory, and Figure 7 is a VSP horizontal feature. FIG. 1...Photoelectric conversion unit, 2...Pattern register,
3...Line width calculation unit, 4...Character frame detection unit, 5...
Division point determination section, 6... Vertical sub-pattern extraction section,
7...Vertical sub-pattern memory, 8...VSP horizontal feature extraction unit, 9...VSP horizontal feature pattern memory, 10...Horizontal sub-pattern extraction unit, 11...
...Horizontal sub-pattern memory, 12...HSP vertical feature extraction section, 13...HSP vertical feature pattern memory, 14...Right diagonal sub-pattern extraction section, 15
...Right diagonal sub-pattern memory, 16...RSP
Left diagonal feature extraction unit, 17...RSP left diagonal feature pattern memory, 18...Left diagonal sub pattern extraction unit, 19...Left diagonal sub pattern memory, 20...
...LSP right diagonal feature extraction unit, 21...LSP right diagonal feature pattern memory, 22...VSP horizontal feature vector extraction unit, 23...HSP vertical feature vector extraction unit, 24...RSP left diagonal feature vector extraction unit,
25... LSP right diagonal feature vector extraction unit, 26...
...Identification unit, 27...Dictionary, 28...Character name output,
31...Control circuit, 32...Change point detection circuit, 3
3... Change point memory, 34... Change point counter, 3
5...Pattern memory address counter, 36...
...Comparator, 37...Arithmetic circuit, 38...Register A, 39...Register B.

Claims (1)

[Claims] 1. Scanning a character figure pattern in a predetermined direction, detecting a cross section of a character line in the scanning direction, and extracting a cross section whose cross section length is sufficiently longer than the character line width of the character figure pattern. Extract sub-patterns in multiple directions by scan in the direction, extract the sum of K powers (K is a constant) of the distances between all character lines on the scanning line on one side from the origin and the origin as the first feature, and The sum of K powers (K is a constant) of the distances between all the character lines on the scanning line on the side and the origin is extracted as a second feature, and the first feature obtained for all the sub-patterns and the A feature extraction method characterized in that the second feature is a character/figure pattern feature.