JPS6013515B2 - Online handwritten character recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a recognition method in which when a character is written with a pen on a sheet of paper placed on a tablet, an electrical signal representing the character recognition is generated in real time.

In the conventional recognition method of this type, in order to remove the bounce that occurs at the stroke end of a character pattern, a polygonal line approximation is performed, and then the length of the line segment at the stroke end is compared with a preset reference value, and the line segment at the stroke end is If it was too short, I would have thought it was a joke and deleted it.

However, this method has the disadvantage that line segments that are relatively short at the ends of the stroke but still contain useful information may be lost.
Therefore, an object of the present invention is to improve the above-mentioned drawbacks of the conventional technology, and its features include storing in a coordinate memory a coordinate string indicating the pen position sampled every short time of handwritten characters drawn on a tablet; Based on the movement speed of the writing point calculated from the coordinate string and the pressure contact information between the pen and paper, continuous writing point movement sections (a,, a2, a3, a
4), and if the end point (p) of the pen point movement section coincides with the end point of the stroke, the length of the pen point movement section on the coordinate plane and the same stroke (A, .A2, A3). When comparing the length of other pen point movement sections and finding that the former is less than a certain percentage, the coordinate string of that section is deleted from the coordinate memory, and the contents of the resulting coordinate memory are used as the target for character feature extraction. There is an online handwriting recognition method like recognition.

Therefore, if the total points of the stroke are the stop points of the writing point (coordinates where the speed does not reach the speed of the writing point movement section), no splash processing is performed on that stroke (that is, the stroke is not deleted). Examples will be described below with reference to the drawings.

FIG. 1 is an example of an input character pattern.

Here, the writing point speed S(Pn) at a certain sample point Pn
is defined as follows. S (Pn) D (PMI) Here, (Pn, PM) is the moving distance of sample points Pn and Pn+ per sample time.

FIG. 2 shows the pen speed determined for all sample points according to this formula and expressed in a graph.

On the graph, an "x" mark indicates a pen-down state, and a "0" mark indicates a pen-up state. Next, among the sample points in the pen-down state, a continuous sample string having a velocity above a certain level is determined and defined as a pen point movement section (a,, a2, a3, a4). In FIG. 2, a to a4 are the pen point movement sections. In a spring pattern like A3 in FIG. 1, the pen speed at the end points is high, so the shape becomes like a3 in FIG. 2. Therefore, on such a graph, when there is a point movement section that is in contact with the pen-up state section (P in Figure 2), that section (
A3) is determined to be a possibility of a splash, and the length of that section on the coordinate plane is compared with the length of the pen point movement section (ra) next to it, and if (a3) is shorter than a certain value, it is determined that there is a splash. from the input coordinate information. If such a pen point movement section does not exist and the stroke is at a stopping point (coordinates where the velocity is below the dotted line in FIG. 2), no splash processing is performed for that stroke. In the example of FIG. 1, the end point p of the first stroke is not the stopping point, so the length of A3 including this point is L(n), and the length of A2 is L(A2), and the predetermined value C( 0.2 to 0.3 degrees), and when obifu <C, A3 is considered to be a fly, and the coordinate information belonging to it is deleted.

The result of deletion of A3 is shown in FIG. Next, a block diagram of a handwritten character pattern splash processing circuit based on the above principle is shown in FIG. 4, and its operation will be explained.

1 is a coordinate memory, 2 is a distance calculator, 3 is a section division point detector, 4 is a section division point registration memory, 5 is a section length calculator,
6 is a register, and 7 is a judge.

The input coordinate data S, is stored in the coordinate memory 1, and at the same time, the distance calculator 2 calculates the distance between sample points, that is, the writing speed at each sample point.

A section dividing point detector 3 detects a point where the writing point speed changes above or below a certain level, and stores an address corresponding to the dividing point on the coordinate memory 1 in a section dividing point registration memory 4.
An interval length calculator 5 calculates the length of each interval by adding the distances between sample points and stores it in a register 6.
When the end point of a pen point movement interval and the stroke end point coincide, the determiner 7 compares the length of that interval with the length of the adjacent interval within the same stroke by reading it from the register 6, and determines whether the former is the same. If it is shorter than a certain value, it is determined that it is unnecessary data due to a splash, and a dummy pen-up state coordinate value is written in the area where the corresponding coordinate data in the coordinate memory is stored, making it invalid information. After completing the processing for all strokes in this manner, feature extraction and character determination are performed by reading the coordinate data S2.

As explained above, in this embodiment, the phenomenon peculiar to splashes, in which the pen point is in a moving state at the end of the stroke, is extracted by calculating the pen speed for each sample point, and the splashes on the character pattern are removed. This method has the advantage that only valid coordinate data can be recognized as a target for feature extraction.

Since the present invention performs splash removal processing on the input pattern, it has the advantage of obtaining high-quality character patterns, and has the advantage of obtaining high recognition rates for real-time handwritten characters such as numbers, alphabets, kana, and other symbols. It can be used for recognition.

[Brief explanation of the drawing]

FIG. 1 is an example of an input pattern, FIG. 2 is a graph representing the writing point speed, FIG. 3 is a diagram of processing results, and FIG. 4 is a block diagram of a splash processing circuit. Length: Coordinate memory, 2: Distance calculator, 3
...Section dividing point detector, 4...Section dividing point registration memory, 5...Section length calculator, 6...
...Register, 7...Judgment device. Figure 1 Figure 2 Figure 3 Figure 4

Claims (1)

[Claims] 1 A coordinate string indicating the pen position sampled at short intervals of handwritten characters drawn on the tablet is stored in the coordinate memory, and the motion speed of the pen point calculated from the coordinate string and information on the pressure contact between the pen and paper are also stored. Then, find a continuous pen point motion section with a speed above a certain level, and if the end point of the pen point motion section coincides with the end point of the stroke, the length of the pen point motion section on the coordinate plane is the same as the end point of the stroke point. When comparing the length of other pen point movement sections in the stroke and finding that the former is less than a certain percentage, the coordinate string of that section is deleted from the coordinate memory, and the contents of the resulting coordinate memory are used as the target for character feature extraction. An online handwritten character recognition method characterized by recognizing characters.