KR100537827B1 - Method for the Separation of text and Image in Scanned Documents using the Distribution of Edges - Google Patents

Abstract

1. TECHNICAL FIELD OF THE INVENTION

The present invention relates to a method for separating a scan image using boundary distribution and a computer-readable recording medium having recorded thereon a program for realizing the method.

2. The technical problem to be solved by the invention

The present invention obtains a background brightness of a scanned image, blocks the scanned image into a predetermined size, and divides the scanned image into a background area, an image area, or a character by using an average brightness and a dispersion of brightness and a boundary line and a boundary line dispersion. It is an object of the present invention to provide a method of separating a scanned image using a boundary distribution for dividing into regions and a computer-readable recording medium recording a program for realizing the method.

3. Summary of Solution to Invention

According to another aspect of the present invention, there is provided a method for separating a scanned image using a boundary line distribution, the method comprising: a first step of receiving an image including an image region and a text region to obtain a background brightness of the image; A second step of blocking the received image into a predetermined size; Obtaining a mean brightness and a dispersion of the brightness and a boundary line and a boundary line dispersion of each blocked block; A fourth step of dividing each block into a background area, an image area, or a text area by using the obtained brightness dispersion, average brightness, and boundary line dispersion; And a fifth step of correcting errors by performing post-processing on the separated background area, image area, and text area.

4. Important uses of the invention

The present invention is used for a multifunction machine, a copier, a scanner, and the like.

Description

Method for the Separation of text and Image in Scanned Documents using the Distribution of Edges}

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a method for performing phase separation of a scanned image using a borderline distribution and a computer readable recording medium having recorded thereon a program for realizing the method. More particularly, the present invention relates to a charge coupled device (CCD) or a contact. A computer-readable recording method for separating the scanned image into a background region, an image region, and a character region by scanning a document image scanned by an image sensor (CIS) or the like and a program for realizing the method. It is about the medium.

Binary processing and compression of documents containing pictures, as in normal document storage systems or fax machines, can lead to serious loss of information. In order to solve this problem, compression to 256 gray-level or color images itself is required. This is less loss of information in the image compared to binary compression, but is greatly reduced in terms of compression efficiency. On the other hand, if the quantization is increased when the image is stored in the Joint Photographic Coding Experts Group (JPEG) method to increase the compression rate, many errors occur in the character portion.

Therefore, in order to efficiently process and compress a document image including an image, the image part is compressed in a gray level or color in the JPEG method, and the character part is a JBIG (Joint Bi-level Image Group) method. Should be compressed. To this end, a method of separating the image area and the text area from the document image is required.

In general, a method using frequency analysis is mainly used as a method of analyzing a given image and dividing it into an image area and a text area. To do this, use a Fast Fourier Transform (FFT) or calculate a system even with neighboring pixels (horizontal or vertical) in a local region of N (horizontal) * M (vertical) pixels. The method using the distribution of is mainly used.

An image having continuous gradation, such as a photograph, has a low frequency characteristic in which the gradation gradually changes due to its characteristics. Therefore, when the continuous grayscale image is divided into N * M local regions and FFT transformed, the value of the self-producing wave is larger than that of the character region. On the contrary, in the case of the character area, the gray level changes abruptly at the boundary of the character stroke, so that the value of the high frequency part appears larger than the continuous gray image area such as a photograph during FFT conversion.

However, FFTs are not suitable for analyzing images in real time because they require quite a lot of mathematical operations due to their computational characteristics. Therefore, many methods are used in place of frequency analysis using even the system, which calculates even the system with neighboring pixels in the local area, and there are many small values close to 0 in the histogram of this value. The image can be distinguished by an image having strong surface low frequency, that is, a continuous gradation image region such as a photograph. On the contrary, a large number of large values exist in the distribution of values even with neighboring pixels and a value of 0 even in the rest of the total distribution indicates If it is occupied, the high frequency is strong, so it is determined by the character area.

In this case, the reason why the value of the text area is 0 is the most part of the entire distribution, because the background of the text is usually white or non-white. This is because the inside of the stroke is a constant color, in most cases black.

With the development of printing technology, binary output images are used in many books, newspapers, magazines, etc., and the outputs of inkjet printers and laser printers spread from house to house are also binary output images. The binary output image is processed by error diffusion or dither screening to process the continuous grayscale image as a network, and therefore, it is seen as a continuous grayscale image to the human eye. Appears strong. Therefore, there is a problem in that the conventional method is not easy to separate the binary image included in the character area.

The present invention has been proposed in order to solve the above problems, by using the average brightness and brightness distribution of each of the blocks obtained by obtaining the background brightness of the scan image and blocking the scan image to a predetermined size by using the boundary and boundary line dispersion SUMMARY OF THE INVENTION An object of the present invention is to provide a method for separating a scanned image using a boundary distribution for separating the scanned image into a background region, an image region, or a character region, and a computer-readable recording medium having recorded thereon a program for realizing the method. Other objects and advantages of the present invention can be understood by the following description, and will be more clearly understood by the embodiments of the present invention. Also, it will be readily appreciated that the objects and advantages of the present invention may be realized by the means and combinations thereof indicated in the claims.

According to an aspect of the present invention, there is provided a method for separating a scanned image using a boundary line distribution, the method comprising: a first step of receiving an image including an image region and a text region to obtain a background brightness of the image; A second step of blocking the received image into a predetermined size; Obtaining a mean brightness and a dispersion of the brightness and a boundary line and a boundary line dispersion of each blocked block; A fourth step of dividing each block into a background area, an image area, or a text area by using the obtained brightness dispersion, average brightness, and boundary line dispersion; And a fifth step of correcting errors by performing post-processing on the separated background area, image area, and text area.

On the other hand, the present invention, in order to separate the upper and lower portions of the scanned image by using a boundary line distribution, to receive the image containing the image region and the character region to the image separation apparatus of the scan image having a processor to receive the background brightness of the image Obtaining a first function; A second function of blocking the input image into a predetermined size; A third function of obtaining an average brightness and a dispersion of brightness and a boundary line and a boundary line dispersion of each blocked block; A fourth function of dividing each block into a background area, an image area, or a text area using the obtained brightness dispersion, average brightness, and boundary line dispersion; And a computer-readable recording medium having recorded thereon a program for realizing a fifth function of correcting errors by performing post-processing on the separated background area, image area, and text area.

The above objects, features and advantages will become more apparent from the following detailed description taken in conjunction with the accompanying drawings, whereby those skilled in the art may easily implement the technical idea of the present invention. There will be. In addition, in describing the present invention, when it is determined that the detailed description of the known technology related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted. Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart illustrating an embodiment of a phase separation method of a scanned image using a borderline distribution according to the present invention.

First, the brightness of the background included in the entire input image is obtained, and the entire input image is divided into blocks having a size of 16 * 16 pixels. Then, the boundary and the variance of the divided blocks, the average and the variance of the brightness are obtained. Thereafter, each block is classified into a background block, an image block, or a character block using this.

Then, the noises included in the classification process, as well as the parts of the image and the text are incorrectly classified by the partial characteristics. To do this, we use contextual judgment. In addition, most of the images included in the document are used in the form of rectangles to make the classified images into rectangles, thereby correcting partially misclassified images so that the boundaries of the images are surely included in the images. To this end, blob coloring is performed on an image in advance, and a different number is assigned to each image. Due to this process, the images included in the document can be extracted separately.

Herein, in more detail, a method of classifying an image including images and characters includes images in which white and black have a continuous tone and characters have a large difference in intensity. Keep crossing. Therefore, the image generally has a small edge and the character has a large boundary at the intersection of white and black.

However, it is difficult to reliably classify an image block and a character block only by the sum or average of the boundary lines in the block. The reason is that the image block has a small boundary of each pixel, but all pixels have a value of some size, whereas the character block has a large boundary at the intersection of white and black, but the range is very narrow, and the rest is white. Because the solid color of, the boundary has a value of 0 or extremely small, and a simple sum does not give a clear difference enough to be used for separation.

It is the dispersion of the boundary line that is suggested as an index that will reflect the boundary line characteristics in the image block and the character block. Dispersion is very small in an image block in which a similar boundary line continues, whereas in the case of a character block in which a boundary line close to zero and a very large portion coexist, dispersion is very large. As shown in Fig. 2, it can be seen that the boundary variance of the image tends to be distributed in a smaller direction than the boundary variance of the character. Accordingly, a block having a larger value than a predetermined value may be classified into a letter, and a block having a small value may be classified into an image. At this time, the criterion should be set slightly larger so that the case where an image has a value larger than a character and is misclassified as a character as little as possible.

However, this method alone can be misclassified. First, when the boundary distribution of an image block is large, it is incorrectly classified as a letter. This mainly corresponds to a block including light and dark boundaries in the image or a block belonging to the boundary between the image and the background. Since these blocks represent a boundary, they are mostly distributed in a line having a thickness of 1 to 2 blocks.

That is, the blocks classified by the letter are included among the blocks classified into the image. However, since the reference value is set high so that the image is not classified into characters at the time of classifying by variance of the boundary line, the number is not large. Therefore, most modifications can be made in context-based Determination.

The problem is that the characters belong when they are smaller than the standard. This occurs when a block contains only a portion of the character, that is, between the character and the margin, between the character and the character, as well as between the stroke and the stroke within the character, and even if the character is included correctly, the brightness of the black is not small enough. Sometimes.

In order to solve this problem, the boundary variance of a block is normalized to a value obtained by inverting the average brightness of the block. Here, the normalization is a value obtained by inverting the average brightness, and as the number of times is divided, the difference between the boundary variance of the image and the boundary variance of the character becomes slightly clear, but the difference is very small. Thus, a single normalization process allows a sufficiently accurate classification.

The cumulative distribution in which the dispersion of the boundary line used when the dispersion of the boundary line is small for each image block and the character block is accumulated is normalized to a value in which the average brightness is inverted. As shown in Fig. 3, it can be seen that the boundary line dispersion value of the character block is very large compared to the value of the image block. By using these characteristics, blocks with small dispersion of boundaries can be easily classified into images and texts.

Here, the value in which the average brightness of the block is inverted means 255 minus the average brightness when a 256 gray-level image of the document is taken as an example.

Next, we will look at how to classify a background in an image that includes images and text.

The background of the document is usually a solid color close to white. Therefore, if the classification is made in consideration of only the boundary line, it is classified as an image because the dispersion value of the boundary line is very small. However, since the background occupies a considerable area of the entire document, it is necessary to classify the background separately because the compression efficiency increases when the binary compression or separation is performed like the character.

To do this, a histogram according to brightness is used for pixels with large brightness (more than 206 in the case of 256 gray-levels) for the document image, taking advantage of the fact that the background is close to white and monochromatic. Is set to the background brightness of a given document image. In the classification process, if the dispersion of the brightness in the block is small enough and the average of the brightness is close to or brighter than the previously obtained background, it is classified as the background before the classification according to the boundary line.

4 is a flowchart illustrating a phase separation process of an input image according to the present invention.

As shown in FIG. 4, when the image is input, the brightness of the background included in the entire input image is obtained, and the entire input image is divided into blocks having a size of 16 * 16 pixels. Then, the boundary and the variance of the divided blocks, the average and the variance of the brightness are obtained. Thereafter, each block is classified into a background block, an image block, or a character block using this.

5 is a flowchart illustrating a process of obtaining a brightness of a background according to the present invention.

As shown in FIG. 5, a histogram of brightness is drawn for only pixels having a brightness greater than 205 among the pixels of the entire image by receiving a document image having 256 gray-levels as an input. . Then, find the brightness (mode) that appears the most from the completed histogram and set it as the brightness of the background.

6 is a flowchart illustrating an embodiment of a process for obtaining a variance of a boundary line according to the present invention.

As shown in FIG. 6, the boundary line used to obtain the boundary line variance is first obtained by receiving a 16 * 16 pixel block as an input, and then the dispersion of the boundary line is obtained. In this process, the average and variance of brightness are additionally calculated. On the other hand, the brightness of all the pixels in the block is not obtained separately. Here, the Sobel Operator, which is the most widely used method, is used to calculate the variance of the boundary line.

Looking more closely at this, there are two boundary lines corresponding to one pixel, horizontal and vertical, and since the stroke of the letter may be horizontal, vertical, or diagonal, it may cover all three directions. Find both the horizontal and vertical boundaries. Therefore, the boundary lines that can be obtained in one block are all 16 * 16 * 2 = 512, and the variance of the boundary lines is obtained from this.

Here, the edge at the coordinates (x, y) is defined as shown in Equation 1 below. only Is the brightness of pixels (x, y).

In addition, the dispersion of the boundary line is defined as in Equation 2 below. Here, if the pixel at which the block starts is a coordinate (0, 0), and the block is located at row j of column i, the pixel at which the block starts in the entire image becomes (j * 16, i * 16).

In addition, the average brightness of the block and the dispersion of the block are defined as shown in Equation 3 below.

7 is a detailed flowchart illustrating a phase separation method of a scanned image using a borderline distribution according to the present invention.

First, when an image including an image area and a text area is input (701), the background brightness of the entire image is obtained (702). Thereafter, the entire image is divided into blocks having a size of 16 * 16 pixels (703). Then, the average brightness and the dispersion of the brightness and the boundary line and boundary line dispersion of each block are obtained (704).

Subsequently, the following classification process is performed using the background brightness of the entire image, the average brightness of each block, the dispersion of the brightness, and the boundary distribution.

First, before the classification by boundary variance, the variance of the brightness in the block is sufficiently small, i.e., less than the predetermined threshold and the average of the brightness of the block is brighter or similar to the brightness of the background (705, 706). Classify (707). On the other hand, if the distribution of brightness in the block is smaller than the predetermined threshold and the average brightness of the block is not brighter than the brightness of the background (705, 706), it is classified as an image area (710).

On the other hand, if the variance of the brightness in the block is greater than the predetermined threshold (705) and the variance of the boundary is greater than the predetermined threshold (708), it is classified as a character area (711). On the other hand, if the variance of the brightness in the block is greater than the predetermined threshold (705) and the variance of the boundary is not greater than the predetermined threshold (708), then the character area is included in the image area. Use the normalized value as the inverted value.

In other words, if the normalized value is greater than a predetermined threshold, the comparison is made (709), and if it is large, it is classified as a character area (711). If not, it is classified as an image area (710).

Through the above process, an image including an image and a text can be easily divided into a background area, an image area, and a text area.

Next, we will look into the post-processing process of contextual judgment, blob coloring, and squareting of pictures.

Blocks classified in the classification stage of the image contain some error, mainly due to noise mixed in the background or text part, and when a part of the bold text is classified as the image, the boundary between light and dark in the image Are mostly classified as letters. These blocks are usually separated by oneself and exist in a line or line. Therefore, in order to correct this, a context-based determinition method of reclassifying itself by referring to what is classified as a neighboring block is used. In this case, since blocks containing errors may exist as lines, each block should be implemented in each direction rather than referring to the blocks in the horizontal direction and the vertical direction at the same time.

Looking at this in more detail, the background block and the character block is represented by 0 and the image block is represented by 1 in the image of which phase separation has been completed. At this time, the type of the block located in row j of column i When = 0 and 1, the contextual judgment in the horizontal direction is given by Equation 4 below.

In addition, the contextual judgment in the vertical direction is as shown in Equation 5 below.

On the other hand, the image which has undergone the above contextual determination process is assigned each consecutive number by blob coloring. Then, the shape of the picture is made into a rectangular shape by using the consecutive numbers given by the blob coloring during the squareization of the picture.

The background area and the text area are displayed as 0 by the blob coloring and the numbers 1, 2, 3 ... are assigned to the picture. Create a space to store the coordinates of the top, bottom, left, and right of each image in advance, and check the numbered blocks one by one, and continuously update the coordinates corresponding to the number each time the image is encountered. It is easy to see. When all the same numbers are assigned to the blocks inside the rectangle composed of the coordinates thus obtained, the work of converting the image into the rectangle is completed. This operation removes the misclassified character blocks that might still exist in the image (actually they may be characters, but if they are characters in the background, it would be right to classify them as images) and ensure that the boundaries of the image are included in the image. You can do that. At this time, it is not a problem in terms of information loss that the background and some of the characters are compressed to 256 gray-levels, but binary compression of the images is a serious problem. Therefore, it is not a big problem that the image is slightly wider while fitting the rectangle.

Additionally, small images can be viewed as errors and reclassified as background, text. This can be easily obtained by using the coordinates of the image obtained above, and comparing this with the area of the entire document image. In one embodiment of the present invention, the image is deleted if the area of the image is smaller than 1/700 of the total area. Of course, even if an image is deleted, the number is kept continuously.

8 is an exemplary view showing a result of performing a post-treatment process after performing the present invention.

After the original document image 81 is separated and separated, the result 83 obtained by finally quadrangling the result 82 through the contextual judgment process, which is a post-processing process, is shown.

As described above, the method of the present invention may be implemented as a program and stored in a recording medium (CD-ROM, RAM, ROM, floppy disk, hard disk, magneto-optical disk, etc.) in a computer-readable form. Since this process can be easily implemented by those skilled in the art will not be described in more detail.

The present invention described above is capable of various substitutions, modifications, and changes without departing from the technical spirit of the present invention for those skilled in the art to which the present invention pertains. It is not limited by the drawings.

As described above, the present invention has the effect of being processed in the same way as the electronic image not only for the electronic input image but also for the scanned image applied to a multifunction apparatus, a copier, a scanner, and the like that support various resolutions.

In addition, the present invention is applied to a multifunction device, a copier, a scanner, and the like, which supports various resolutions, to not only read a specific image area among image areas, but also to apply an algorithm suitable for the characteristics of the separated area to improve image quality. There is an effect that can increase the processing speed.

In addition, the present invention has an effect that can be applied to an embedded system with low computational capability to improve the performance of the embedded system by separating the station by simple operation as soon as possible.

1 is a flowchart illustrating an embodiment of a phase separation method of a scanned image using a borderline distribution according to the present invention;

2 is a cumulative distribution graph of boundary variances for images and characters according to the present invention;

3 is a cumulative distribution graph of values obtained by normalizing the variance of the boundary lines for images and characters according to the present invention with a value inverting the average brightness;

4 is a flow chart of an embodiment of a phase separation process of an input image according to the present invention;

5 is a flowchart illustrating a process of obtaining a brightness of a background according to the present invention;

6 is a flow chart of an embodiment of a process for obtaining a variance of a boundary line according to the present invention;

7 is a detailed flowchart illustrating a phase separation method of a scanned image using a borderline distribution according to the present invention;

8 is an exemplary view showing a result of performing a post-treatment process after performing the present invention.

* Explanation of symbols for the main parts of the drawings

21: Cumulative distribution graph of boundary variances for an image

22: cumulative distribution graph of boundary variances for characters

31: Cumulative distribution graph normalizing boundary variance for an image

32: Cumulative distribution graph normalizing the boundary variance for characters

Claims (4)

In the phase separation method of the scanned image using the boundary distribution, A first step of receiving an image including an image area and a text area to obtain a background brightness of the image; A second step of blocking the received image into a predetermined size; Obtaining a mean brightness and a dispersion of the brightness and a boundary line and a boundary line dispersion of each blocked block; A fourth step of dividing each block into a background area, an image area, or a text area by using the obtained brightness dispersion, average brightness, and boundary line dispersion; And A fifth step of correcting an error by performing a post-processing process on the separated background area, image area, and text area Phase separation method of the scanned image using a boundary distribution including a. The method of claim 1, The fourth step, A sixth step of determining the block as the background area when the brightness variance of the block obtained in the third step is smaller than the first threshold and the average brightness of the block is greater than a value obtained by subtracting a predetermined value from the brightness of the background; A seventh step of determining the block as an image area when the brightness variance of the block obtained in the third step is smaller than the first threshold and the average brightness of the block is not greater than a value obtained by subtracting a predetermined value from the brightness of the background. ; An eighth step of determining the block as a character area when the brightness variance of the block obtained in the third step is not smaller than the first threshold and the boundary variance of the block is larger than the second threshold; The boundary variance of the block is inverted to the average brightness of the block without the brightness variance of the block obtained in the third step being less than the first threshold and the boundary variance of the block not greater than the second threshold. A ninth step of discriminating the block as an image area when the normalized value is not larger than a third threshold; And The boundary variance of the block is inverted to the average brightness of the block without the brightness variance of the block obtained in the third step being less than the first threshold and the boundary variance of the block not greater than the second threshold. A tenth step of determining the block as a character area when the normalized value is larger than the third threshold Phase separation method of the scanned image using a boundary distribution including a. In order to separate the upper and lower portions of the scanned image by using the boundary distribution, in the reverse separation apparatus of the scanned image having a processor, A first function of receiving an image including an image area and a text area to obtain a background brightness of the image; A second function of blocking the input image into a predetermined size; A third function of obtaining an average brightness and a dispersion of brightness and a boundary line and a boundary line dispersion of each blocked block; A fourth function of dividing each block into a background area, an image area, or a text area using the obtained brightness dispersion, average brightness, and boundary line dispersion; And A fifth function of correcting errors by performing post-processing on the separated background area, image area, and text area A computer-readable recording medium having recorded thereon a program for realizing this. The method of claim 3, wherein The fourth function is, A sixth function of discriminating the block as a background area when the brightness variance of the block obtained by the third function is smaller than a first threshold and the average brightness of the block is greater than a value obtained by subtracting a predetermined value from the brightness of the background; A seventh function of discriminating the block as an image area when the brightness distribution of the block obtained by the third function is smaller than the first threshold and the average brightness of the block is not greater than a value obtained by subtracting a predetermined value from the brightness of the background; ; An eighth function of discriminating the block as a text area when the brightness variance of the block obtained by the third function is not smaller than the first threshold and the boundary variance of the block is larger than a second threshold; The brightness distribution of the block obtained by the third function is not smaller than the first threshold, and the boundary distribution of the block is not greater than the second threshold, and the boundary distribution of the block is a value inverted the average brightness of the block. A ninth function of discriminating the block as an image area when the normalized value is not larger than a third threshold; And The brightness distribution of the block obtained by the third function is not smaller than the first threshold, and the boundary distribution of the block is not greater than the second threshold, and the boundary distribution of the block is a value inverted the average brightness of the block. A tenth function of discriminating the block as a text area when a normalized value is larger than the third threshold A computer-readable recording medium having recorded thereon a program for realizing this.