JPH056461A - Noise eliminating system - Google Patents

Abstract

PURPOSE:To quickly perform the picture noise elimination processing by eliminating a relatively large noise and keeping end point forms against character blur. CONSTITUTION:Picture dot data A of character '6' which is read from a form document and is segmented is compressed to a half longitudinally and transversely to obtain data B. A maximum picture element block 32 is extracted from this compressed data to obtain data D by exclusion of the other components. This maximum picture element block 32 is enlarged to the original picture element side to obtain data E, and AND between this data E and original picture data A is operated. Thus, even relatively large noises 22 to 24 are not recognized as character components but are eliminated. The number of processing picture elements is reduced by picture compression to increase the processing speed. Finally, AND between processed picture data and original picture data is operated, and original character blur parts or the like are restored as they are to keep forms of end points.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image data processing method, and more particularly to a noise removing method for removing noise which is harmful to character recognition from image data read by a character recognition device.

[0002]

2. Description of the Related Art One of data input means for a data processing device
As a general rule, OCR (Optical Character Recogni)
A character recognition device called “tion” is often used. This device electrically recognizes a printed or handwritten character based on an input from a reading sensor such as a photoelectric conversion element, and allows simpler and faster input than keyboard input. In this apparatus, a large number of standard patterns are registered in advance, and pattern matching is performed with the characteristic patterns extracted from the read characters to extract and output some approximate candidate characters.

When a document is read by such a character recognition device, image noise (hereinafter simply referred to as noise) together with the original image information of the character is caused by dust or dirt on the document or the reading sensor section. Is often read. Although this noise hinders character recognition after reading, it needs to be removed, but conventionally, a method has been adopted in which the surrounding state is checked for each dot of the read image information to make a determination.

[0004]

By the way, in recent years,
There is an increasing demand for improvement in processing performance for such a character recognition device, and the performance of the reading sensor, the image memory, and the like is improved in cost-to-cost ratio, so that the resolution of reading the document is increasing. As a result, since the number of pixels to be processed has also increased, the conventional method for making noise determination for each dot requires a huge amount of time as a whole, and there is a problem in that the processing speed is reduced. Further, when the noise is comparatively large, it is not regarded as noise and is recognized as image information forming a character, which cannot be removed.

On the other hand, when the density of the characters on the original is partially low and, for example, the end portions of the characters (hereinafter referred to as end points) are faint, the faint parts are regarded as noise. It is often removed, and the shape of the end point changes from the original character. In order to avoid this, it is necessary to perform noise removal processing for each dot by a complicated algorithm, which causes a problem that the processing speed further decreases and the processing program becomes large.

Therefore, there is a problem that the above problems must be solved.

The present invention has been made in order to solve the above problems, and it is possible to remove a relatively large amount of noise, while maintaining the shape of the end point for faint characters and performing high-speed noise removal. The purpose is to obtain a noise removal method that can achieve

[0008]

A noise removing method according to the present invention is an image processing apparatus for performing various image processing on image data read from an original and cut out,
Compress the cut out image data to a predetermined magnification, (ii) extract the maximum pixel block from the compressed image data,
(iii) After expanding the extracted pixel block to the original size again, (iv) noise removal is performed by taking the logical product of the expanded image data and the original cut-out image data.

[0009]

In the noise removing method according to the present invention, the image data that has been cut out is once subjected to image compression, so that the presence of noise can be clarified as noise and character fading can be integrated into the original character portion. .. Further, by extracting the largest pixel block from the image data after compression, it is possible to remove noise whose existence is clarified. Furthermore, by taking the logical product of the image data after noise removal and the original cut-out image data, only the original character constituent part can be restored.

[0010]

EXAMPLES The present invention will be described in detail below with reference to examples.

FIG. 1 shows a character recognition apparatus using a noise removing method according to an embodiment of the present invention. As shown in this figure, the character recognition device 11 is provided with a reading unit 12 so that the characters on the form 13 can be read as image information. The reading unit 12 is connected to the noise removing unit 15 via the character cutting unit 14. This character cutout 1
The character 4 is designed to cut out a character from the image information input from the reading unit 12 on a character-by-character basis and send this to the noise removing unit 15.

The noise removing section 15 removes the noise component from the image data corresponding to one character cut out by the procedure described in detail below, and sends it to the feature extracting section 16. The feature extraction unit 16 extracts a feature pattern from the image data from which noise has been removed, and the pattern matching unit 1
Output to 7. The pattern matching unit 17 that received this
Is distance data indicating the degree of approximation of each of the extracted feature patterns and a large number of standard patterns stored in advance in the dictionary storage unit 18, and a predetermined number of characters are calculated in order from the smallest value. Is output as a candidate.

Next, the operation of the noise removing section 15 will be described in detail with reference to FIG. Here, 1 from "0" to "9"
Digit numbers are read, and of these, for example, "6"
The case of reading is described with reference to FIG.

The image data of the character "6" read by the reading unit 12 from the form 13 and cut out by the character cutting unit 14 is temporarily stored in a buffer memory or the like (not shown) (step S101 in FIG. 2). ). Here, FIG.
As shown in (A), the extracted character image data is
In addition to the original character component 21, noise components 22 to 24 are included.

Next, the noise removing section 15 compresses the cut-out character image data to 1 / n in both vertical and horizontal directions and stores it in another buffer memory (not shown), as shown in FIG. 3B. It is stored (step S102, FIG. 3B). here,
n is a positive integer. Now, for example, when n is “2”, the area of 2 × 2 dots is compressed and converted into 1 × 1 dot. In this case, if any one of the four dots is black as shown in FIG. 4A, one compressed dot is made black as shown in FIG. 4B. On the other hand, when all four dots are white as shown in FIG. 5A, one dot after compression is made white as shown in FIG. Thereby, for example, the area of 8 × 8 dots as shown in FIG. 6A is compressed into the area of 4 × 4 dots as shown in FIG. 6B.

Next, the noise removing section 15 operates in step S10.
The maximum pixel block is extracted from the compressed data compressed in 2 and stored in the buffer memory (step S103 in FIG. 2, FIG. 3C). The extraction process in this case is performed as follows. That is, as shown in FIG. 7A, first, the horizontal projection of the target image data is obtained.
In this case, the horizontal projection 2 corresponds to the noise components 22 and 23.
6, a horizontal projection 27 corresponding to the character portion 21 and a horizontal projection 28 corresponding to the noise component 24 are obtained. Similarly, the vertical projection 29 is obtained. Then, by ANDing the horizontal projections 26 to 28 and the vertical projection 29, three pixel blocks 31 to 33 in which black dots are concentrated are obtained. Then, the largest pixel block 32 is selected and extracted from these areas, and as shown in FIG. 7B, other pixel blocks containing a noise component are excluded.

In this way, the compressed data (see FIG.
When the largest pixel block is extracted from (C)) ((D) in the same figure), the noise removing unit 15 again performs processing to expand it to the original number of pixels (step S104, FIG. 3).
(E)). At this time, one black dot as shown in FIG. 8A is enlarged to a black 2 × 2 dot as shown in FIG. 8B. On the other hand, one white dot shown in FIG. 9A is enlarged to a white 2 × 2 dot shown in FIG. 9B. As a result, for example, as shown in FIG.
The area of × 4 dots is enlarged to the area of 8 × 8 dots shown in FIG. The image data obtained in this way is a slightly thick and sharp character image with no noticeable roughness although the roughness is conspicuous.

Next, the noise removing section 15 ANDs the enlarged pixel data (FIG. 3 (E)) with the original clipped image data (FIG. 3 (A)) (step S10).
5). As a result, the noise components 22 to 24 are removed from the original clipped image data, and only the character component 21 is output as it is. At this time, the shapes of the end points of the character are reproduced as they are, including the blurred character.

When image compression is performed in step S102, if a character faint dot 35 as shown in FIG. 11A exists in a part of the original cut-out image data, it is compressed. By doing so, a compressed image in which dots are continuous (FIG. 7B) is obtained. Similarly, supposing that a character fading dot 37 as shown in FIG. 12A exists in a part of the original cut-out image data,
By compressing this, a compressed image in which dots are continuous (FIG. 7B) is obtained. In these cases, step S
Even if the maximum block extraction processing in 103 is performed, the compressed dots 36 and 38 are not removed. Therefore, the original character fading dot 3
5, 37 are reproduced.

On the other hand, as shown in FIG. 13A, when the dot 39 as a noise component exists, the compressed image as shown in FIG. 13B is obtained by compression. In this case,
Since the compressed dots 40 are removed by the maximum block extraction processing in step S103, the original noise component is not reproduced by the AND processing in step S105.

FIG. 14 shows how the image data shown in FIG. 13A is processed in association with FIG. As shown in this figure, the original cut-out image data (A) is compressed in half vertically and horizontally (B), and the dots 40 are removed by maximum block extraction (C).
Then, this is enlarged again to become (D), and the original image data (A) and AND are taken to become (E).

In this embodiment, an example in which a 2 × 2 dot area is compressed into a 1 × 1 dot area has been described.
In this case, the number of dots is reduced to 1/4, so that the overall processing speed is improved. Furthermore, if 4 × 4 or 8 × 8 dots are compressed to 1 × 1 dot, further improvement in processing speed can be expected.

In the present embodiment, the character recognition device has been described as an example, but the present invention is not limited to this and can be widely applied to image processing devices.

[0024]

As described above, according to the present invention,
Since it is not determined whether or not each dot is noise as in the conventional case, and determination is made in pixel block units, even relatively large noise can be removed without being recognized as a character component. Further, since the maximum pixel block is extracted after compressing the image data, the number of pixels can be reduced, and the processing speed can be improved.

Further, since the logical product of the noise-removed image data and the original cut-out image data is taken, the original character fading portion and the like are restored as they are, and the shape of the end point can be maintained. There is also an effect.

[Brief description of drawings]

FIG. 1 is a block diagram showing a character recognition device using a noise removal method according to the present invention.

FIG. 2 is a flow chart for explaining the contents of this noise removal method.

FIG. 3 is an explanatory diagram showing an example of noise removal by this noise removal method.

FIG. 4 is an explanatory diagram showing an example of an operation at the time of image compression.

FIG. 5 is an explanatory diagram showing an example of an operation at the time of image compression.

FIG. 6 is an explanatory diagram showing an example of an operation at the time of image compression.

FIG. 7 is an explanatory diagram for explaining an operation at the time of extracting a maximum pixel block.

FIG. 8 is an explanatory diagram illustrating an example of an operation when enlarging an image.

FIG. 9 is an explanatory diagram illustrating an example of an operation when enlarging an image.

FIG. 10 is an explanatory diagram illustrating an example of an operation when enlarging an image.

FIG. 11 is an explanatory diagram showing an example of an operation at the time of image compression.

FIG. 12 is an explanatory diagram showing an example of an operation at the time of image compression.

FIG. 13 is an explanatory diagram illustrating an example of an operation at the time of image compression.

FIG. 14 is an explanatory diagram showing an example of noise removal by this noise removal method.

[Explanation of symbols]

 11 Character Recognition Device 12 Reading Unit 13 Form 14 Character Extraction Unit 15 Noise Removal Unit 16 Feature Extraction Unit 17 Pattern Matching Unit 18 Dictionary Storage Unit 21 Character Component 22-24 Noise Component 32 Maximum Pixel Block

Claims (1)

Claim: What is claimed is: 1. An image processing apparatus for performing various kinds of image processing on image data read from an original and cut out,
The cut-out image data is compressed to a predetermined magnification, the maximum pixel block is extracted from the compressed image data,
A noise removal method characterized in that the extracted pixel block is enlarged again to its original size, and then the product of the extracted pixel block and the cut-out image data is ANDed to remove noise.