JPS603071A - Character recognition system - Google Patents

Abstract

PURPOSE:To improve the precision of character recognition by scanning a subpattern vertically or horizontally, and raising the distance between a character frame side and a character line to the (N)th power-sum of a normalized value. CONSTITUTION:A subpattern extraction part 4 scans vertically on the entire surface about pattern registers to extract a vertical subpattern (VSP) on the basis of the relation between the continuous length of a black dot and line width calculated by a line width calculation part 3. Similarly, a horizontal subpattern (HSP) is extracted by a horizontal scan, a right slanting subpattern (RSP) is extracted by a right slanting 45 deg. scan, and a left slanting subpattern (LSP) is extracted by a left slanting 45 deg. scan. Then, the feature extraction part 6 starts making a horizontal scan with the vertical subpattern at the point above the left vertical side among four sides constituting the character frame to detect all change points from a white point to a black point.

Description

DETAILED DESCRIPTION OF THE INVENTION (Technical Field) The present invention relates to a fast and accurate feature extraction method.

(Background technology) In conventional character/figure recognition devices, strokes are extracted from the character □□□ shape number turns, and then recognized using the position, length, mutual relationship between strokes, etc. of the extracted strokes. Many methods have been adopted.

The equipment in this building (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 large curvature values as branching points; (2) Extract the strokes by combining the series, or (2) perform thinning processing on the character/figure pattern to create a skeleton, trace the connectivity of the skeleton pattern and trace the skeletal pattern, and detect sudden angular changes, etc. It detected and extracted 10 strokes, extracted geometrical rib features, etc. from the extracted strokes, and recognized character shapes. However, (11 methods (・ko1,
When the character (figure pattern) becomes large and the character figure pattern becomes double-acting 6, the amount of processing increases and the processing speed decreases.In method (2), it is necessary to make the character figure pattern thinner. Further, there are problems such as pattern distortion and whiskers due to the thinning of the lines, and the subsequent processing becomes complicated.

(Problems to be solved by the invention) The purpose of the present invention is 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 of a character figure pattern. The position of the entire character line on the scanning line is detected by scanning from the point , and the distance between the scanning start point and the character line (1 is 1
3iJ Normalized by the size of the character circumscribing frame in the scanning direction -C
l'llll tlj '1-Divide the side-by-side array of the N-th power sum! 1-1IG (-\I) whose objective is to extract a small group, and to provide a fast and stable character recognition device if7+.

(Structure of the invention J-5 to σ J-11) Fig. 1 shows an embodiment of the character recognition device of the present invention.
Figure 3 shows an example of Sanobakun, and Figure 3 shows an example of erectile drawing Lfj.
.

In Fig. 1, 1 (Kako Report 1 Shunkan 1711 (, 2 is the pattern register, ; U is the line width word mouth" + 1 part, Roto Zabu pattern drawing In Bti, 5 (Kako letter frame + 1゜old') 1shi, 6(1
, Edge Distance Calculation Department, 7('', Heavy Equipment Matrix Sen II Notification'i
ll, )N, l: Identification i'if), 0 is character name output -〇-.

The operation of this embodiment is that the characters on the book closing set in the reading mechanism are converted into binary quantized tickle electricity and digit symbols by photoelectric conversion>xl+ +, and the pattern register 2
is stored in At the same time, the line width (W) of the input cover pattern is calculated in the line width calculating section 346. Sub-pattern extraction unit 4 (Sat, the pattern register is vertically scanned over the entire surface, and the continuous length of black hints and line width division unit; ~1S11J).Similarly, the horizontal sub pattern (H4F) is extracted by horizontal scanning.
A right diagonal subpattern (l(, SP
), the left center first sub-pattern (L,'-Ef) is extracted by scanning 45 degrees diagonally to the left.

Figure 2 is an example of the original pattern and each sub-pattern, where (21) is the original pattern and (1)) is the vertical sub-pattern (VSI').
, (C) is the horizontal subpattern (ll5P ), ((1)
is a right diagonal sub pattern (R8P), (cl is a left diagonal sub pattern:,' (LSP).Character frame detection unit 5
is a circumscribed 1- to the character/figure pattern in the pattern register.
A rectangular frame (hereinafter referred to as a character frame) is detected, and position coordinates for defining the character frame on a two-dimensional plane defined by a pattern register are sent to the feature extraction unit 6.

In the following description, a coordinate system will be used in which the left-hand side of the character frame is the origin, the horizontal direction is the X axis, and the vertical direction is the Y axis.

The feature extraction unit 6 first starts horizontal evaluation of the vertical sub-pattern from point P(0, y) on the left side -1-, which is the vertical side among the four sides constituting the character frame, and Detect the change points from a dot to a black dot, and calculate the distance between the detected change point 11 and the point 1 on the vertical side where the scanning started [L square X coordinate difference] The length of the character frame in the X direction is normalized using a normalization constant, and the division of the value of X to the Nth power (N is a constant, in this example, ゛ is N-2) is calculated as follows: 1) IJ was detected. - Do this for the change point at 8 (・, the sum is arrayed ■ p, (y)
is stored in ``1-''.However, the interior point represents the back jIt part of the character, and the black point represents the 9. capital in the character '.Also, formula (1)
is the formula for the above-mentioned Vpy), and △XA is the distance between the change point and the scanning start point on the side of the character frame, 〃-1,..., 1\ , 1 (represents the number of detected change points. Also, ΔX in equation (1) is the water s of the character frame.
(The length in one direction is k), and C is an integer constant k). In this example, (1, C=5 (+) is used.

Perform the process similar to above (2) starting from all points on the two vertical sides of the character frame, and create the vertical sub pattern by starting horizontal scanning from the point on the left side of the character frame. Extract the array Vr(i) created by starting horizontal scanning from the point "-" on the right side of the character frame.However, 1-09 ・,YT; This is the Y coordinate of the top side.By similar processing, vertical scanning is performed from all points on the two horizontal sides of the character frame for the water 31z sub pattern, right diagonal sub pattern, and left diagonal sub pattern. Then, for the horizontal sub pattern, the array ``h(j),i
+ zu (''), ℃・te arrangement R for the right diagonal sub-pattern,
! , (j), HbJ), array Lh(J) for the left diagonal diagonal pattern, and extract (1) below.

SeiJ-0,..., Xlt, The subscript represents a point on the horizontal bottom side of the character frame as the starting point, or a point on the horizontal top side as the starting point.
l also h (J), H so, (j).

When extracting Ll, fj) and Lx(j), the vertical length ΔY of the character frame was used as a normalization constant for the distance between the start point of scan 15 and the change point. Character mark 1
．． IJ Kusu extraction part 7 is! Time extraction '*' rlS Using the array of 8 rivers extracted in 6, divide each array into M pieces (He4 is a constant, M1 in this example, He4-7),
By calculating J1 the average value of the array values within the same division 71j of the divided array, a 1×8-dimensional six-character matrix P CIn 1]) is extracted7, where n =,
, 1 ,..., to 4. n=l,...-28 identification>511
8 is a standard character mask f (m,
n) between which distance (1)) shown in equation (2) is M
Calculate the distance ν11 (or 1) and select the smaller value.Output the category name of the standard character mask as a character name ()-\.

D-~t'7TTT seat, 1])-' (m animals) (2) (Effect of emitted jllJ) J to L As explained, 4) Signal mark l-1) Kusu extraction part of this example Features extracted in (1) The marks represent the position, length, direction, etc. of the stroke of the character graphic pattern, and express the characteristics typical of the character 71. or,
As observed in Figure 3, an example of a literary pattern with similar two solitary forms and a graphical representation of the extraction arrangement using the feature extraction unit, local differences in characters are sufficiently reflected in the arrangement. Is it recognized because it is? Be able to measure the direction of 4 degrees. In addition, when creating each array in the feature extraction section, the distance between the point on the side of the character frame where scanning started and the literary line is normalized by the size of the character frame in the scanning direction, and the handrail characters are It is possible to absorb the fluctuations in font size due to differences in writers, which is unique to 46 characters. In addition, since feature extraction from text/figure patterns is achieved through simple scanning and processing, advanced recognition or DJ capabilities are possible, and the equipment is small! 11! ! It also has the advantage of being able to be used in a variety of ways.

The present invention scans a character figure pattern (1:) vertically or horizontally with a sub pattern from which stroke components are extracted in each direction, and calculates the distance between the character frame side and the character line by 1 [1] of the character frame in the scanning direction. The sum of the N-th power of the values normalized by the size is 44. Since city signs are used as signs, complex processing is required.Also, since city signs are stably extracted by following the deformation of handwritten characters, it can be used in high-speed and highly accurate character recognition devices.

[Brief explanation of the drawing]

Figure 1 shows an example of the letter 1 of the present invention, an example of the character 1; The figure shows the features and examples step by step.]...Light weight/weight conversion unit 2...Pattern register; 3
Line width drop section 1... Sub pattern extraction section 5... Character frame structure i, L'lj'tB (i... Feature extraction section 7...
・71 Force signature matrix YuderiB 8...Identification part 9・
...Character name output! Time d71 output Sunoki TU: Ki Kogyo Co., Ltd. Special 671 Output 1 Ruyo”! [Single-person dialect Buri Megumi Yamaki - Ticket / Illustration 2 Figure < o, ) < b no niri (d) < e)

Claims (1)

[Claims] A subpattern is created by scanning a character/figure pattern in a desired 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. Extraction is performed in multiple directions, and for each extracted subpattern, m
Scanning in a predetermined direction using the edge point of the character circumscribing frame of the "l" written figure pattern as a starting point, detecting the positions of all character lines on the scanning line, and the point on the edge where scanning started,
The process of extracting the sum of powers (N is a constant) of the value obtained by normalizing the distance to the detected character line by the size of the Ail character circumscribing frame in the predetermined direction is carried out for each of the subpatterns, and Starting from all points on at least two of the four sides of the character circumscribing frame, scan lines are scanned as a unit, and for each sub-pattern, each scanning line is scanned using the sides of the character circumscribing frame as a unit. The array of the N-th power sum of 1llll L13 L, the extracted N-th power sum array is each divided into M pieces (M is a constant), and the average value of the N-th power sum within each division part is calculated as follows: By extracting M-dimensional feature vectors from each array, the character figure (turn)
- A character recognition method characterized in that a character is recognized by extracting a group of feature vectors and comparing them with a dictionary expressed in the same format as the extracted il# feature vector group.