JPS5852783A - Feature extraction system - Google Patents

Abstract

PURPOSE:To perform character recognition accurately and speedily by extracting segments forming the contour of a character pattern part hided behind a character pattern part, and using them as character features. CONSTITUTION:A character pattern is scanned horizontally or vertically and segments forming the contour of a character pattern are extracted in the form of a set of black-white change points Q1-Q4 between the background and character and used as character features for character recognition. Further, a segment l6 forming the contour of a character pattern part hided behind a character pattern part appearing firstly in the horizontal or vertical scanning direction is extracted in the form of a set of odd-numbered black-white change points Q3... except the 1st change point, and this is used as character features.

Description

[Detailed Description of the Invention] The present invention relates to a character feature extraction method used for bright line character recognition.
This method attempts to enable feature extraction of parts that are hidden by conventional methods.

In character-texture, the characteristics of characters viewed from the left and the characteristics viewed from above are determined and used as data for character recognition. For example, taking the number 3 shown in Figure 1, the handwritten cough number 3 is scanned with a line scanner to obtain a binary signal, which is temporarily stored in memory, and the video signal of number 3 is cut out from the mosquito memory. Generally, it is scanned together with other handwritten characters and stored in the memory)
, first find a collection of points that convert from 0 (background) to 1 (character) on each horizontal scan from left to right, and use this as the contour feature of the mosquito number 3 seen from the left. .

In this logic, line segment'! 1 r 14e AB is the feature of the character seen from the left (referred to as the left feature in ζ\).

Similarly, if we find the group of points that first change from 0 to 1 on each vertical scanning line scanning from top to bottom, we get -14* in Figure 2.
z= is obtained, which becomes the characteristic feature (upper feature) of the character when viewed from above. These left features, upper features, right % features, and lower features obtained in the same way are easy to extract, and can be easily extracted from simple patterns such as the numbers Ω, 1, 2, etc. It becomes good feature data for the character ga, and can be effectively used for the character ga. For example, there is a line segment 4 with a person between the line segments t1 and t3 above, which is 6 in the numbers 0 to 9.
Therefore, since the target is any number from 0 to 9, it is possible to recognize the number 3 with just this feature alone. However, there are many features that are hidden in the background and are not extracted. For example, in the left feature, the large concave part P1
*P! etc. are not extracted, and in the upper part, only the part that is a roof or eaves is extracted, and everything below it is hidden and not extracted. Therefore, the feature extraction is insufficient for characters with slightly complex patterns such as kanji and kana, and some characters cannot be separated or identified from other characters.

Therefore, the present invention aims to make accurate and quick character & I recognition possible by making it possible to extract features of hidden parts. That is, a Honbunmeisha character pattern is scanned horizontally or vertically, line segments forming the outline of the character pattern are extracted as a result of black-and-white transition points between the background and the characters, and these are used as character features for the characters. In the percentage feature extraction method, Tani Mizuyuki, the character mark hidden behind the character pattern part that first appears in the scanning direction is defined as a collection of odd-numbered qIt-th black-and-white transition points in the vertical scanning direction, excluding the first one. The present invention is characterized in that a line segment forming an outline of /S is extracted and used as the character feature, which will be described in detail below with reference to the drawings.

The line segment LlpL*·-m- is recognized as follows. That is, scanning line Zl) e zly 4 has different Y coordinates, so if we find the first point 111 (X 411) that causes 0→1 inversion for each Y coordinate, we get (Y6 *O)+
(Yi +0)t(Y鵞pO) v (Ys exs
) e (Y41X4') A data group such as "" is obtained. Here, 0 indicates ``There is no 0→1011 reversal point'', but such a thing can also be simply expressed as Xl(i=o, 1
, 2...), the above data is (Xi,
Yl). These data are stored in memory
If you take the method of recording the X number in the i-address,
Easy to process. That is, the address counter is sequentially +1 and y, l y, l yj...... Because of the address data, xl, x, ...... are successively generated, and ΔX=X
When calculating l+1-Xi, the line segment t1e z,
14, the difference ΔX is small, so it can be determined from this that it is a continuous line segment. Also, if jX<0, it is a line segment going down the left shoulder, and if jX>0, it is a line segment going down the right shoulder.

At the 1.0 boundary between line segment t1 and ZI F L, jx suddenly becomes large. In such a case, it can be said that the linearity is interrupted, or at least connected by a line segment parallel to the horizontal scanning line. This discontinuity point and X
This is the point where it has a certain value from l-0. line segment t
The starting end of 1 is the latter, and the ending is the former, the line segment t1 has the former at both ends, and the starting end of line segment t3 is the former, and the ending is the latter. The fact that both ends are discontinuous points, such as the line segment depth, means that for numbers 0 to 9, which are connected into one, there are character pattern portions at both ends, and both ends are closed. Line segment t1. It can be said that both sides of ts are open. Having a line segment with both ends closed is a major feature of the number 3, as mentioned above.

By the way, the number S has a concave PlpP snow, so if this is also detected, the feature of the number 6 will be extracted even better. This concave part could not be extracted because it is in the shadow of the line segment Zl e LM, but the extraction logic is not the "first 0 → 1 change point" but the third (or odd numbered point if generalized). ) 0 → 1 change point", then fr:, the part can be extracted.

That is, as shown in FIG. 6, the scanning line t passing through the concave portion P1
Looking at the 0-1 reversal for aK, it is Qt t Ql
There are four points, #Ql and Q4, and the reversal point Q3, which defines the outline of the shaded end portion t6, is the sixth. The line segment t@n14 can be extracted using the logic of this "third reversal point," and the line segment L- obtained using the method shown in Figure 1, that is, the logic of the "first reversal point"
If you use 1 minute t8, which is 0 or \, you will get 1 minute 4 that has entered the deepest part of the recess. In other words, when we find this fraction of jX, it changes from positive to negative to positive to negative, which clearly represents the characteristics of the number 6. Combining the line segment t- and the line segment zt t Zl, that is, the line segments exist in the order of L*eLseLs in the vertical direction, and on the right of the line segment 4 t z in the horizontal direction) both ends According to the logic that there is a 1/4 that overlaps with the line segment zt e z, even if it is handwritten very roughly, the number 6
can be separated and identified from others.

FIG. 4 shows an example of a handwritten "chi", and FIG. 5 shows an example of a handwritten "te". It is not uncommon for these differences to be determined by the presence or absence of the protrusion R. However, this is hidden in the upper character pattern part B, so
It cannot be extracted using the logic of the first change point in the diagram. On the other hand, if we use the logic of the third change point, especially the third change point in the vertical scanning line, we can extract the line segment 1m * zl (both ends are extracted at the "first change point"). . Line segment 2. If +21· can be extracted, the presence or absence of the protrusion R can be checked by checking whether the change in the Y coordinate is uniform (the vertical direction is X), and it becomes possible to identify the tip and the tip. There are many characters that are confusing, for example, the katakana tsuto chikara, ritotsu, mi and shi, riton, and niton, etc., are randomly written and cannot be distinguished by simple left-hand or upper-hand features. It is expected that these difficult-to-identify breasts will be able to be identified by extracting the chest area using the "odd number" logic, or by combining it with a simple left-hand feature or the like.

In the recognition of handwritten numbers and kana, a large number of features, such as 1000 types, are used, and these constitute a tree circuit of about 50 stages, and character recognition is performed by running through the tree circuits. The features extracted according to the present invention are also used in addition to one of them. Note that if the features are appropriate, results can be obtained by missing a relatively small number of rows, and character recognition speed can be increased. In this respect, the method of the present invention is extremely effective.

[Brief explanation of the drawing]

1 and 2 are explanatory diagrams of the conventional method, and FIGS. 6 to 5 are explanatory diagrams of the method of the present invention. 't1-t in the drawing
1. is a line segment, and zs t t7 e zs e Zl. is a line segment of a hidden character pattern part. Applicant Fujitsu Ltd. Representative Patent Attorney Minoru Aoyagi Figure 1 Figure 2 Figure 3 Figure 4 Figure S

Claims (1)

[Claims] A character feature in which a character pattern is scanned horizontally or vertically, line segments forming the outline of the character pattern are extracted as a collection of black and white transition points between the background and characters, and the line segments are used for character recognition. In the feature extraction method, the outline of the character pattern part hidden behind the character pattern part that appears first in the scanning direction is defined as a collection of odd-numbered black-and-white transition points other than the first in each horizontal and vertical scanning line direction. A character feature extraction method used for character recognition characterized by extracting a line segment and using it as the character feature.