JPS6053919B2 - Real-time handwritten character recognition method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a real-time handwritten character recognition device that recognizes characters, symbols, and numbers input from a tablet.

It is generally well known that in real-time handwritten character recognition devices, unlike OCR and the like, stroke count and stroke order information are added as new recognition information, and recognition can be performed at a relatively high recognition rate. Conventionally, information on the number of strokes in this new recognition information has been used for dictionary selection, and the number of strokes is first extracted from information such as input characters, and matching is performed only with category dictionaries that correspond to this number of strokes. This allows for highly accurate recognition.

However, this recognition method has completely different category dictionaries with different number of strokes. I ended up ending up with τ on the rt day as well. Therefore, in the past, this problem was dealt with by creating additionally registered dictionaries called separate stroke number dictionaries by allowing a wider range of stroke number fluctuations, but this had the disadvantage of significantly increasing the dictionary's capacity. .

Therefore, the purpose of the present invention is to provide a novel real-time handwritten character recognition device that eliminates the above-mentioned drawbacks.In a real-time handwritten character recognition system that uses a relative position matrix representing the relative positional relationship of representative points as a feature, A stroke combining circuit is provided in advance to virtually connect strokes so that the strokes of the input characters are continuous, and the features of the virtually connected input characters are extracted by the stroke combining circuit to generate the input characters. This can be achieved using a real-time handwritten character recognition method that is characterized by recognizing

The present invention will be explained below with reference to the drawings. FIG. 1 shows an embodiment of the real-time handwritten character recognition method of the present invention. In the figure, 1 is a tablet, 2 is a preprocessing circuit, 3 is a stroke buffer, 4 is a stroke combination circuit, 5 is a total distance measurement circuit, 6 is a representative point extraction circuit, 7 is a relative potential matrix creation circuit, and 8 is matching 9 indicates a circuit, and 9 indicates a dictionary.

For example, if "thread" is input from the tablet 1, the input character information is input to the stroke buffer 3 via the preprocessing circuit 2, and the number of strokes and each coordinate value of the stroke are input. (See FIG. 3, 1) The stroke combining circuit 4 combines the input strokes according to the stroke order as shown in FIG. 3, 2. The output from the stroke combination circuit 4 is input to the total distance measuring circuit 5, which determines the total stroke distance and the number of representative points determined from this distance. Based on the determined number of representative points, each representative point is extracted by a representative point extracting circuit as shown in FIG. 3.

A relative position matrix is created by the relative position matrix creation circuit 7 from the coordinates of the representative point determined in this way, and the dictionary 9 and matching circuit 8 sequentially compare it with the contents of the dictionary to extract the answer. The configuration other than the stroke coupling circuit shown in the embodiment of FIG. It is detailed in No. 147083 and Japanese Patent Application No. Sho M-145675.

In the present invention, a stroke combination circuit is provided so that the relative position matrix can be used as a feature even for characters with a large number of strokes. Next, FIG. 2 shows an embodiment of the stroke coupling circuit according to the present invention. In the figure, 10 is a stroke number comparison circuit, 11 is a stroke counter, 12 is a stroke readout circuit, 13 is a stroke buffer All4 is a stroke end detection circuit, 15 is a stroke write circuit, 16 is a stroke buffer 13B117 is a coincidence output,
18 indicates stroke information. The stroke buffer 13 stores coordinate values sent from the preprocessing circuit. (A signal instructing the stroke end for each stroke is also stored.) Similarly, the stroke number comparison circuit 10 sequentially counts the number of strokes from the preprocessing circuit and compares it with the count from the stroke counter 11. . Coordinate values are sequentially read out from the stroke buffer 13 and stored in the stroke readout circuit 12. Ru. The end of the stroke is detected from the coordinate values sent from the stroke reading circuit 12, a count-up signal is output to the stroke counter, and the read coordinate values are sent to the stroke writing circuit 15.

The stroke counter 11 counts the count-up signal from the stroke end detection circuit 14 and is reset by the coincidence output 17 from the stroke number comparison circuit 10.

The stroke comparison circuit 10 compares the number of strokes sent from the preprocessing circuit with the number of strokes from the stroke counter, and outputs a coincidence signal 17 if they match.

Further, the stroke writing circuit 15 sequentially and continuously stores the coordinate values sent via the stroke end detection circuit 14 in the stroke buffer 16.

Thereafter, coordinate data is sent to the total distance measuring circuit along with the matching circuit output. In this way, in the stroke coupling circuit, by sequentially storing the coordinate values corresponding to the strokes as shown in FIG. 2a and the coordinate values sequentially as shown in FIG. 3. This creates a state in which the strokes shown in FIG. 2 are combined. In this way, the present invention combines input strokes in a continuous form in order to enable the use of a feature matching method called a relative position matrix that does not depend on the number of strokes, so the features stored in the dictionary are relatively small. It is possible to at least increase the recognition rate.

[Brief explanation of drawings]

FIG. 1 shows an embodiment of the real-time handwritten character recognition device of the present invention.
FIG. 2 shows an embodiment of the stroke coupling circuit according to the present invention.
FIG. 3 shows the processing status of each part in FIG. 1. In the figure, 1 is a tablet, 2 is a preprocessing circuit, 3 is a stroke buffer, 4 is a stroke combination circuit, 5 is a total distance measurement circuit, 6 is a representative extraction circuit, 7 is a relative position matrix creation circuit, 8 is a matching circuit, 9 is a dictionary, 10 is a stroke number comparison circuit, 11 is a stroke counter, 12 is a stroke reading circuit, 13 is a stroke battler, 14 is a stroke end detection circuit, 15 is a stroke writing circuit, 16 is a stroke buffer, and 17 is a coincidence output. , 18
indicates stroke information, respectively.

Claims (1)

[Claims] 1 In a real-time handwritten character recognition method that uses a relative position matrix representing the relative positional relationship of representative points as a feature, a stroke combination circuit that virtually connects strokes in advance so that the strokes of input characters are continuous. 1. A real-time handwritten character recognition system, characterized in that the stroke combination circuit extracts the features of input characters virtually combined to recognize the input characters.