JPS61121184A - Character recognizer - Google Patents

Abstract

PURPOSE:To obtain accurately the features about the flexion properties by extracting the flexion properties detected at the back of a character pattern and collating the flexion property data with the flexion property data on a standard pattern. CONSTITUTION:For extraction of features, a CPU11 control a series of processing for detection of the flexion properties and stores a character pattern converted into a label to a memory 12. Here a labelling circuit 13 converts the picture element of a black area into a label in terms of the boundary between the black and white areas of the character pattern. Then this labelled character pattern is scanned by a scan circuit 14. When a label having the prescribed direction and position is contained in a label adjacent to the label of the detected picture element, the coordinate position of the detected label is tracked. Thus the depressed part of the picture element is detected, and the data on said depressed part is collated with the data on the flexion properties of each character included previously to the data on a standard pattern. Then the character pattern is recognized.

Description

[Detailed Description of the Invention] <Technical Field of the Invention> This invention reads patterns of characters, figures, symbols, etc. (hereinafter collectively referred to as "characters") by optical means, etc. After digitizing and obtaining the input image, the feature extraction section extracts the geometric features of the character pattern, and the matching section compares the features with standard patterns stored in the dictionary section to recognize unknown characters. The present invention relates to a character recognition device that performs character recognition.

<Summary of the Invention> In the character recognition device of the present invention, the feature extraction unit is provided with a means for detecting the curvature of the second part existing behind the first part of the character pattern, and the matching unit is provided with a means for detecting the curvature of the second part existing behind the first part of the character pattern. It includes a means to perform a comparison based on gender data, which makes it possible to accurately distinguish multiple characters that cannot be distinguished by characteristics other than curvature.

<Background of the Invention> Generally, in a character recognition device, as shown in FIG.
The characters and the like displayed above are optically read by the reading section 2, and the character pattern is converted into digital information by the A/D converter 3, then converted into black and white binary information, and stored in the image memory 4. The next preprocessing circuit 5 inputs the data in the image memory 4 and performs so-called preprocessing such as noise removal, smoothing, and stringing, and the next feature extraction circuit 6 inputs the data in the image memory 4. Extract the geometric features of a character pattern. The dictionary matching circuit 7 includes the above-mentioned feature extraction circuit 6.
The features extracted by It executes recognition processing, identifies one character from among the candidate characters, and outputs that character as a result.

For example, when a character pattern for the kana character "shi" as shown in Figure 11 is input, the characteristics of this pattern such as the number of connected components and curved parts are used to select "shi" as a candidate character. , "mi", and "tsu" are obtained. Next, a character stroke as shown in FIG. 12 is obtained by using the four-direction approximate stroke shown in FIG. 13, and one character among the candidate characters is specified from this character stroke,
This allows it to recognize unknown characters.

However, when a character pattern as shown in Fig. 14 is input, the character stroke becomes as shown in Fig. 15, and in this case, it is difficult to specify it as either the kana character "shi" or "tsu". It's impossible. In the conventional device, as shown in Fig. 16, from one observation frame side 9,
Scanning is performed sequentially toward the frame side 10 opposite to this,
Changes in the distance from observation frame side 9 to the character pattern based on the position where the scanning line meets the character pattern (indicated by in the figure)
This is due to the fact that the curvature of the character pattern is determined based on this detection. That is, the conventional device cannot detect the curvature existing behind a part of the character pattern as shown by G in FIG. 14, but can detect the curvature in the part shown by h in FIG. 16. However, in cases where the curvature of the G part is an important distinguishing factor, as in the case of "shi" and "tsu", the disadvantage is that it can result in misreading or jikuto (unreadable). was there.

<Object of the invention> The present invention is intended to overcome the above-mentioned drawbacks, and provides a character recognition device that can accurately incorporate features related to curvature, thereby reducing misreading and illegibility. The purpose is to

<Structure and Effects of the Invention> In order to achieve the above object, the character recognition device according to the present invention has a feature extraction unit detect the curvature of the second part of the character pattern that exists behind the first part of the character pattern. In addition to providing a curvature detection means for detecting by boundary tracking,
The matching section includes a curvature data matching means for comparing the detected curvature data with the curvature data in the standard pattern.

According to the present invention, it is possible to reliably capture information regarding the curvature of a character pattern, thereby providing a character recognition device that is less likely to be read incorrectly or unreadably, and achieves the remarkable effects of achieving the object of the invention.

<Description of Embodiments> FIG. 2 shows an example of the configuration of the curvature detection means included in the feature extraction circuit 6. As shown in FIG. Of these, CPU (Central
Processing Unit ) 11 controls a series of processes related to curvature detection, and also controls image memory 12
stores a character pattern as shown in FIG.

The labeling circuit 13 focuses on the boundary between the black background and the white background (outer boundary) of the character pattern, and labels the pixels of the black background part (character part) under the following conditions.
Further, the scanning circuit 14 is used to sequentially scan the labeled character pattern. RAMl5 is the above CP
This is a memory that stores programs for U11 and various data for work. The curvature data matching means in this invention is included in the dictionary matching circuit 7, and is also included in the C
It can be configured using PU and RAM.

Next, with reference to the flowchart shown in FIG. 1, the operation of this embodiment will be explained by taking as an example the detection of a recess, which is one aspect of curvature.

First, in step 21, the entire surface of the character pattern shown in FIG. Labeling is performed depending on the relative relationship between the pixel and the adjacent pixels, that is, whether the adjacent pixel is on a black background or a white background. Figure 4 shows an example of this labeling. When the pixels adjacent to the left, right, top, and bottom are white, an arcuate mark is placed on the corresponding side of the square representing that pixel. It is being carried out.

Furthermore, for example, pixels whose right and bottom sides are both white are labeled as shown in Figure 5, and as a result, the labeled character pattern as shown in Figure 3 is created. obtain. The upper, lower, left, and right arcuate labels in the illustrated example will be referred to as (u
), (d), <1>, and (r), and for example, the labels in FIG. 5 are shown as (d, r).

When such labeling is completed, in the next step 22, scanning is performed in the direction indicated by 31 in FIG. Detect pixels of.

Considering that the outer boundary of the character pattern goes around the perimeter, there is at least one pixel labeled (d, r), and unless an error occurs, this step 22
detection is always achieved. In the example of Figure 3, pixel 32
corresponds to this. In this way, when the pixel corresponding to (d, r) is detected, in the next step 23,
Tracking is performed starting from this pixel and going counterclockwise around the outer circumferential boundary of this pattern portion. Here, "counterclockwise tracking" means, as shown in FIG. 6(A)-0,
Label of the pixel where the tracking point is located at that point (6th
In the figure, paying attention to the diagonally shaded arcuate) Al to D1, in the case of label A1, the adjacent labels K, A
2-Sequentially detecting whether or not a label having the direction and position of A4 exists according to the order of A2-A4, and if so, moving the tracking point to the coordinate position of the detected label. say.

For example, if we focus on the right-facing label of the label for pixel 32 in Figure 3, there is no A2 type label adjacent to this label, but an A3 type label is adjacent to it above. The tracking point is moved to pixel 33 labeled A3.

The same applies to FIGS. 6('B) to 9, and B2 to B4. C2
~c4. Label detection and tracking point migration are performed in each order of D2 to D4.

In this way, when the "counterclockwise tracking" in step 23 of FIG.
If not (t), the process returns to step 23 to continue tracking the left turn. In the example of FIG. 3, the label 34 is also
Figure 7 (a).

In the example of (B), the labels 41 and 46 are (l), respectively, and when these are detected, the process proceeds to step 25.

Since the outer boundary of the character pattern goes around once, (1
) exists, and the transition to step 25 is always performed. In step 25, counterclockwise tracking is further performed, and in step 26, it is determined whether the label of the tracking point is (r). If the label is not (ri), the process moves to step 28, and it is determined whether the label is (d). If it is not (d), the process returns to step 25 to continue tracking counterclockwise. (d), the label 47 in FIG. 7(B)
As shown in , label (r) is detected without detecting label (r).
This means that d) has been detected, and the process ends with the result that no recess exists in this pattern.

In step 26, the label (r), for example label 3 in FIG.
5 or the label 42 in FIG. 7 (f) is detected,
Since there is a portion with walls on the left and right, that is, a recess (G in FIG. 3), this is detected in step 27, and the process ends.

The data regarding the concave portion, which is one aspect of the curvature of the pattern detected in this way, is compared with data regarding the curvature of each character that has been included in the data regarding the standard pattern, for example, as shown in Fig. 3. It is recognized that the pattern is "shi".

Therefore, in addition to the above example, it is also possible to distinguish between "n" and "n", such as "nu" shown in FIG.
Even in characters with minute protrusions, by detecting the recesses g, it is possible to accurately distinguish between "su" and "nu".

Incidentally, in the above embodiment, tracking was performed in a counterclockwise direction, but clockwise tracking may also be performed as shown in FIG. 9. In that case, step 22 in FIG.
d, A'), label (r) in step 24, label (1) in step 26, etc. In addition, the form of curvature is not limited to concave parts, but can be modified to suit other curvatures such as convex parts, and by tracking in a predetermined order, a set of pixels specific to the curvature can be All you have to do is detect it.

Further, tracking does not need to be performed over the entire circumference of the pattern, but only on the necessary portions. It is desirable to detect curvature after narrowing down the candidate characters.
It may be detected from the beginning. Furthermore, in the predetermined direction, that is, in the above example, the description focused on the curvature of the other part behind one part of the pattern with respect to the scanning direction.
It is also possible to judge the curvature of other parts at the same time.

[Brief explanation of drawings]

FIG. 1 is a flowchart showing a curvature detection process in an embodiment of the present invention, FIG. 2 is a block diagram of a configuration example of a curvature detection means in an embodiment of the invention, and FIG. 3 shows labeled characters. Diagrams showing the pattern, Figures 4 and 5
The figure shows an example of labeling, Figure 6 is a diagram to explain counterclockwise tracking, Figure 7 is a diagram to explain recess detection, and Figure 8 is an example of a character pattern with a recess. diagram showing,
Figure 9 is a diagram for explaining an example of clockwise tracking, Figure 10
The figure is a block diagram showing the general configuration of a character recognition device.
Figure 1 is a diagram showing an example of a character pattern, Figure 12 is a diagram showing a character stroke of the pattern in Figure 11, Figure 13 is a diagram showing an approximate stroke with directionality, and Figure 14 is a diagram showing a recessed part. FIG. 15 is a diagram showing character strokes of the pattern in FIG. 14, and FIG.
FIG. 4 is a diagram showing a conventional recess detection method for the pattern of FIG. 4; 2...Reading unit 6...Feature extraction circuit 7...Dictionary matching circuit 11...C
PU13...Labeling circuit, 14...Scanning circuit 15...R
AM Patent Applicant: Tateishi Electric Co., Ltd. (B) 4 country label ° Ruihi Haiku I IIl Elutriation agency, TB → C diagram, left j, hsuno i3 yori, self-written BR diagram ++2 Figure 701 row

Claims (2)

[Claims] (1) Read the character pattern of the unknown character, convert it into black and white to obtain an input image, and after extracting the geometric features of the input image by the feature extraction section, the matching section extracts the geometric features of the input image. The character recognition device recognizes the unknown character by comparing the output with a standard pattern stored in a dictionary section, wherein the feature extraction section detects a second part existing behind the first part of the character pattern. It includes a curvature detection means for detecting curvature by tracing the boundaries of a character pattern, and the collation unit compares the curvature-related data detected by the curvature detection means with the curvature data included in the standard pattern. A character recognition device comprising curved data matching means for matching. (2) The curvature detecting means includes a means for labeling boundary pixels in a relative relationship with adjacent pixels, a means for tracking the labeled pixels according to a predetermined order, and a means for labeling the boundary pixels in a relative relationship with adjacent pixels, and a means for tracking the labeled pixels in a predetermined order, and said second by sensing a set of pixels.
2. The character recognition device according to claim 1, further comprising means for detecting curvature of the portion.