JPH05210763A - Automatic learning type character recognizing device - Google Patents

Abstract

PURPOSE:To automatically improve a recognition ratio by automatically executing relearning at the time of generating defective recognition in a character recognizing device. CONSTITUTION:The character recognizing device is provided with a camera 1 for inputting a character image, an image memory 2 for storing the fetched image, a character segmenting part 3 for segmenting character data for one character from the memory 2, a neural network 4 for receiving the character data from the segmenting part 3 and selecting a character, a character judging part 5 for judging a character based upon an output value from the neural network 4, a recognition-disabled character storing part 6 for storing the correct character data of a character whose recognition is disabled, and an automatic learning part 7 for automatically changing load data 8 in the neural network 4 by relearning the character data stored in the storing part 6.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention is used in a character recognition apparatus for automatically recognizing an input character, and more particularly to an automatic learning type character recognition apparatus for automatically recognizing an unrecognized character and correctly recognizing it.

[0002]

2. Description of the Related Art As shown in FIG. 3, a conventional character recognition apparatus includes a camera 1 for capturing an image, an image memory 2 for storing the captured image, and a character cutout unit for cutting out one character from the image in the image memory 2. 3, a neural network 4 that receives character data and selects a character, a character determination unit 5 that determines a character from an output value of the neural network 4, and an offline learning unit 9 that sets weight data 8 in the neural network 4. It has and.

Here, the image captured by the camera 1 is stored in the image memory 2 and the image is extracted by the character cutting section 3.
Character data for a character is cut out and sent to the neural network 4. Next, the neural network 4
Is selected from the input character data by using the weight data 8 preset by the offline learning unit 9, and the character determination unit 5 determines the output from the neural network 4 to perform character recognition. ..

[0004]

In this conventional character recognition apparatus, since the weight data of the neural network is set by the offline learning section, in order to improve the recognition rate, the recognition failure character is taken offline and re-read. There is a drawback that it is necessary to learn and it is difficult to automatically improve the recognition rate.

It is an object of the present invention to provide an automatic learning type character recognition device which can automatically improve the recognition rate by eliminating the above-mentioned drawbacks.

[0006]

According to the present invention, a camera for capturing an image, an image memory connected to the camera for storing image data, a character cutting section for cutting character data from the image memory, and the character cutting section. In a character recognition device equipped with a neural network for receiving character data and selecting characters based on load data, and a character determination unit for performing character determination based on the output value of the neural network, a character that has failed recognition in the character determination unit. A recognition failure character storage unit for inputting and storing correct character data of data, and an automatic learning unit for automatically learning the character data stored in the recognition failure character storage unit and changing the load data of the neural network are provided. It is characterized by

[0007]

With respect to the character data which is determined to be recognition failure by the character determination unit, the correct character data is input and stored in the recognition failure character storage unit. Then, the automatic learning unit automatically re-learns the stored correct character and changes the weight data of the neural network.

Therefore, in the neural network, since character selection is always performed with highly accurate weight data, it is possible to improve the character recognition rate.

[0009]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention.

In this embodiment, a camera 1 for capturing an image,
Image memory 2 connected to camera 1 for storing image data
A character cutout unit 3 that cuts out character data from the image memory 2, a neural network 4 that receives character data from the character cutout unit 3 and selects a character based on the load data 8, and a character that determines a character from the output value of the neural network 4. In the character recognition device including the determination unit 5,
Characterized by the present invention are a character recognition unit 5 for inputting and storing correct character data of character data which has been recognized by the character determining unit 5, and character data stored in the character storage unit 6 for recognition failure. An automatic learning unit 7 that automatically learns and changes the weight data 8 of the neural network 4 is provided.

Next, the operation of this embodiment will be described with reference to the flow chart shown in FIG.

The character image captured by the camera 1 in step S1 is stored in the image memory 2 (step S1).
2) The character cutting unit 3 cuts out character data for one character and sends it to the neural network 4 and the character determination unit 5 (step S3). Next, the neural network 4 selects a character using the weight data 8 (step S4), and the character determination unit 5 determines a character from the output value of the neural network 4 (step S4).
5). Then, if it is recognizable, the character is recognized (step S6), and if a recognition failure occurs, when a person inputs a correct recognition failure character from a keyboard or the like, the character data is stored in the recognition failure character storage unit 6. Then, the automatic learning unit 7 re-learns the character data in the recognition failure character storage unit 6 (step S8), and automatically changes the weight data 8 in the neural network 4 (step S9).

As a result, the character selection in step S4 is correctly performed, and the recognition rate can be improved.

[0015]

As described above, according to the present invention,
By providing the recognition failure character storage section and the automatic learning section, since the weight data of the neural network is automatically changed every time recognition failure occurs, the recognition rate can be automatically improved. The effect is great.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is a flowchart showing the operation.

FIG. 3 is a block diagram showing a conventional example.

[Explanation of symbols]

 1 camera 2 image memory 3 character cut-out part 4 neural network 5 character determination part 6 recognition failure character storage part 7 automatic learning part 8 load data 9 offline learning part S1 to S9 steps

Claims (1)

[Claims] 1. A camera for capturing an image, an image memory connected to the camera for storing image data, a character cutout unit for cutting out character data from the image memory, and character data received from the character cutout unit for receiving load data. In a character recognition device comprising a neural network for character selection and a character determination unit for performing character determination based on the output value of the neural network, correct character data of character data that is not recognized by the character determination unit is input and stored. A self-learning type including: an unrecognized character storage unit for automatically learning the character data stored in the unrecognized character storage unit; and an automatic learning unit for automatically changing the weight data of the neural network. Character recognizer.