JP2003250046A - Image processing apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image processing apparatus for efficiently specifying a density of a background by obtaining a density histogram whose frequency of incidence is maximum for every line from a received image and using a FIFO of one line to eliminate the density of the background. <P>SOLUTION: A density histogram is produced from a multi-value document image 101 in the unit of lines in the main scanning direction (step S103), a density value whose frequency of incidence is maximum by one line is searched from the density histogram (step S104), and background is eliminated by using a density conversion expression for one line with respect to pixels distributed to the line on the basis of the density information with the maximum frequency of incidence by each line (step S105). The processing above is executed in all the main scanning directions to output a document image 102 from which the background is eliminated. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing apparatus such as document copy, document conversion, embedded image processing equipment,
In particular, the present invention relates to an image processing device that removes the background.

[0002]

2. Description of the Related Art With the progress of digitization of documents, the character information written on paper is often stored as image information. However, since image information such as a bitmap has a large capacity and it is difficult to edit the document, it has been desired to convert it into a character code. Therefore, OCR (Optical Character Rea) that reads character images and converts them into character codes
der: optical reader) is used. In order to correctly convert the character code into OCR, it is necessary to correctly separate the character area and the background area by the binarization processing performed in OCR.

However, an image having a high background density value, such as a newspaper, may not be correctly separated into characters and background pixels by the binarization process performed by the OCR, resulting in a lower OCR recognition rate. To do. Therefore, it is necessary to remove the background in order to improve the OCR recognition rate for an image having a particularly high density value of the background as a previous process before applying the OCR.

As a conventional method of an algorithm for removing the background, Japanese Patent Laid-Open No. 7-262348, "Method and apparatus for removing background color in image processing", Japanese Patent No. 28866.
There is "Background Removal Method of Image Storage Device" in Japanese Patent No. 93, and "Image Processing Device" in Japanese Patent No. 3134292. These methods use the density histogram of the vertical axis frequency and the horizontal axis density value as the feature amount of the background removal processing.

A general background removal apparatus for improving the OCR recognition rate shown in FIG. 9A is a scanner unit 901 that digitizes and reads character information written on a paper, and image information captured by the scanner unit. And an image storage unit 902 that stores image information after image processing, and a signal processing unit 903 that binarizes an input multivalued digital image.

Arrows in FIG. 9 show the flow of image information. The multi-valued image read by the scanner unit 901 is temporarily stored in the image storage unit 902. Then, the signal processing unit 903 detects the density value of the background and performs background removal processing on the captured multi-valued image. As a result of this background removal process, the image is stored again in the image storage unit 902. After storing the image after the background removal processing in the image storage unit 902, the OCR is performed.
904 receives the image processed by the signal processing unit 903 as input and performs character recognition processing. Since binarization is performed by the OCR 104 using a simple threshold value, the density value of the character area is hardly changed, and the background area of which the density value is low and the background is removed is sent to the OCR 904 as a character recognition target image.

[0007] In addition, in an embedded image processing apparatus that requires real-time processing, it is required to reduce the amount of memory used and an algorithm that pursues image processing speed. It is possible to reduce the number of times of scanning an image as much as possible and store the information of one line without storing the image information in the image memory and then performing the sequential processing.
There is a demand for an algorithm that is compatible with removing the background by using FO to store information for several lines.

FIG. 9B shows the structure of a background removing apparatus for a general built-in real-time image processing apparatus.
The difference from FIG. 9A is that the image data input from the image reading unit 905 is processed by the signal processing unit 903 on a pixel-by-pixel basis or a line-by-line basis. Instead, the image data and the feature amount are stored by using the image storage FIFO 906 for one line capable of storing the image data for one line, the signal processing unit 903 performs the image processing, and the image output unit 907 processes the processed image. Outputting data.

[0009]

However, since character recognition is desired in OCR, a background removal algorithm that improves the recognition rate is desired. The present invention has been made in view of the above circumstances, and an object thereof is to provide an algorithm capable of realizing real-time image processing and an image processing apparatus capable of reducing the amount of storage memory used for processing.

Further, by obtaining a density histogram having the maximum appearance frequency for each line from the input image, it is possible to efficiently specify the density value of the background and remove the density value of the background by using the FIFO for one line. It is an object of the present invention to provide an image processing device that can use the.

Further, the background density value is the density value of the input image for the processed image in which the background is removed by smoothing the density histogram having the maximum appearance frequency for each line. It is an object of the present invention to provide an image processing apparatus capable of using a FIFO for several lines so as to reflect and make uniform.

Further, by normalizing the density value before obtaining the density histogram in which the appearance frequency of each line is the maximum, an image having a high density value as a whole has pixels of an intermediate density value increased, An object of the present invention is to provide an image processing apparatus that can use a FIFO for one line to easily identify a background density value for subsequent processing and efficiently remove the background.

Further, by using the density histogram having the maximum appearance frequency for each line and the density value of the input image,
By changing the density conversion formula for removing the background, the output image has a small change in the density value with respect to the character pixels, and the effect of removing the background with respect to the background pixels is high.
It is an object of the present invention to provide an image processing device that can use a line FIFO.

[0014]

In order to achieve the above object, an image processing apparatus according to a first aspect of the present invention comprises an image input section for digitizing and capturing character information, an image signal processing section for performing image processing calculation, and In a system having a storage unit for storing the processing result, a means for creating a density histogram for each line in the main scanning direction, a means for searching for a density value having the maximum appearance frequency for each line in the main scanning direction, And a means for converting the density value of the image based on the density value information having the maximum appearance frequency for each line in the scanning direction.

According to a second aspect of the present invention, in the first aspect of the present invention, the appearance for each line for several lines obtained by the means for obtaining the density value having the maximum appearance frequency for each line in the main scanning direction. It is characterized in that it has means for smoothing the density value having the maximum frequency.

Further, in the invention according to claim 3, in the invention according to claim 1, means for normalizing the density value is provided before the means for obtaining the density value having the maximum appearance frequency for each line in the main scanning direction. It is characterized by having.

According to the invention of claim 4, in the invention of claim 1, the means for converting the density value of the image based on the density value information having the maximum appearance frequency for each line in the main scanning direction It is characterized in that it has means for changing the means for converting using the density information.

[0018]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows a flow of background removal processing according to claim 1 performed by a signal processing unit 903 of FIG. Figure 1
The density histogram is created for each line in the main scanning direction for the multi-valued document image 101 to be input in the processing order (step S103), and the density value with the highest appearance frequency for each line is searched from this density histogram ( Step S
104) is performed, and based on the density information of the maximum frequency for each line, the background distribution is removed for the pixels distributed in that line by the density conversion formula of one line (step S105). The above-described processing is performed in all main scanning directions to output the document image 102 from which the background is removed.

First, the multi-valued document image 101 is searched for density values for one line in the main scanning direction, and a density histogram is created (step S103) to create a density histogram for one line. Thereby, the density distribution information of one line in the main scanning direction is stored.

Next, the density histogram of one line in the main scanning is searched for the density value with the maximum appearance frequency for each line (step S104). The density histogram of one line in the main scan obtained in the creation of the density histogram (step S103) is searched for, and the density value with the maximum appearance frequency of one line is stored. This value is used in the density conversion formula of the next process. If you save the density value with the maximum appearance frequency of one line,
Background removal by one-line density conversion formula (step S10
Perform 5) to remove the background. The density value with the highest appearance frequency of one line and the minimum density value of the entire image in all main scanning directions are searched before this process, and the density of the pixel of interest of the input image is used by using these two values. Convert using the value in the conversion formula. For pixels belonging to one main scanning direction, the density value with the highest appearance frequency of the line is used, and the density value with the highest appearance frequency for each line in all the main scanning directions and the minimum density value for the entire image are used. The density value is converted for one line to remove the background. When the density value conversion processing is completed for each line, a density histogram for each line is created again for the next line (step S103), and a density value with the highest appearance frequency for each line is searched (step S104), one line. The background removal is performed by the density conversion formula (step S105). When all line processing is completed, the background removal processing is completed and the processed image is output. Since the density value of the pixel in the character area is low before the processing, the density value does not change so much, but the density value of the pixel in the background area becomes low by the conversion formula.

FIG. 2 shows an example of a document image before background removal and an image after background removal. FIG. 2 is an image obtained by cutting out a part of a newspaper read by a scanner. In the upper part (a), the newspaper originally has a high density value in the background. In contrast, Figure 2
In the lower diagram (b), the character area does not change much even if the background density value is effectively converted to a low value.

FIG. 3 shows a density histogram of the images of FIGS. 2 (a) and 2 (b). As is clear from both of these histograms, in (a), the number of pixels having a large distribution around the intermediate density value of 100 as the background is 25%.
It is shifting around. Further, for the pixels having a density value of 150 or more in (a), there is almost no change in the density value in (b).

FIG. 4 shows a flow of background removal processing according to claim 2 performed by the signal processing unit 903 of FIG. In FIG. 4, a density histogram is created (step S103) for the multi-valued document image 101 to be input in the processing order, and based on the density histogram information, a density value having the maximum appearance frequency for each line is searched (step S104). ). Here, it is checked whether or not the appearance frequency of each line can search the maximum density value for the line used for smoothing in the next process. If it has not been saved yet, the image data of the line so far is F
Save to IFO and search for next line. When the line part used for smoothing can be searched and saved, the density value having the maximum appearance frequency in the line is smoothed based on the density value having the maximum appearance frequency for each one of the several lines. .
Based on the maximum density value information, the smoothed appearance frequency of each line is subjected to the background removal 105 by the density conversion formula for the pixels distributed in the line, and the document image 102 from which the background is removed is output.

First, the multi-valued document image 101 is searched for a density value for one line in the main scanning direction, and a density histogram is created (step S103) to create a density histogram for one line. Thereby, the density distribution information of one line in the main scanning direction is stored. Then, the density histogram of one line in the main scanning is searched for the density value having the highest appearance frequency for each line (step S104). The density histogram of one main scanning line obtained in the density histogram creation (step S103) is searched for, and 1
The density value with the highest line appearance frequency is saved. here,
Since the density value information with the maximum appearance frequency for several lines is required to smooth the density value with the maximum appearance frequency of one line, it is checked whether or not the information for several lines is stored. If not stored yet, the next main scan is searched for, and a density histogram is created (step S103) to find the density value with the highest appearance frequency of the line.
This density histogram is repeatedly created (step S103), and if the density value information with the maximum appearance frequency for several lines necessary for smoothing can be saved, the density value information with the maximum appearance frequency for several lines is used. , The density value with the maximum appearance frequency on the line is smoothed (step S201).

Specifically, as shown in FIG. 5, the frequency value a of interest is smoothed by taking the average value of its neighborhood. In the case of smoothing in the vicinity of 3 as shown in FIG. 5, if the frequency values before and after are b and c, respectively, pay attention to the value obtained by averaging the frequency amount a of interest (b + c) / 2. The smoothing process of the maximum density value of the appearance frequency that replaces the existing frequency amount a is performed. After the smoothing of the density value having the maximum appearance frequency on the line (step S201) is finished, the background is removed by the density conversion formula for one line (step S105) to remove the background. The density value with the highest appearance frequency of one line and the minimum density value of the entire image in all main scanning directions are searched before this processing, and the density value of the target pixel of the input image is used by using these two values. Is converted into the conversion formula. For pixels belonging to one main scanning direction, the density value with the highest appearance frequency of the line is used, and the density value with the highest appearance frequency for each line in all the main scanning directions and the minimum density value for the entire image are used. The density value is converted for one line to remove the background. When the density value conversion processing is completed for each line, a density histogram for each line is created again for the next line (step S103), and a density value with the highest appearance frequency for each line is searched (step S104), one line. The background removal is performed by the density conversion formula (step S105). When all line processing is completed, the background removal processing is completed and the processed image is output. Since the density value of the pixel in the character area is low before the processing, the density value does not change so much, but the density value of the pixel in the background area decreases due to the conversion formula.

FIG. 6 shows a result image with and without smoothing processing. Without smoothing processing (a), for each line,
Since the density conversion formula may vary greatly depending on the density value of the maximum frequency, an unnatural edge may occur at the center of the image. On the other hand, in the smoothing process (b), since the average value in the vicinity of the line is taken, the density conversion formulas for the lines are not significantly different, and thus an unnatural edge is less likely to occur in the main scanning direction.

FIG. 7 shows a flow of background removal processing according to claim 3 performed by the signal processing unit 903 of FIG. The density value is normalized (step S701) with respect to the multi-valued document image 101 input in the processing order of FIG. 7, the minimum density value of the input image is normalized to 0, and then the density is increased line by line in the main scanning direction. Create a histogram (step S103),
From this density histogram, the density value with the maximum appearance frequency for each line is searched (step S104). Further, based on the density information of the maximum frequency for each line, the background distribution is removed from the pixels distributed in that line by the density conversion formula of one line (step S105). The above-described processing is performed in all main scanning directions to output the document image 102 from which the background is removed.

Since the input image has a high background density value of the background and a lower density value of the character area, the density histogram has a small difference between the background area density value and the character area density value, and the overall width is narrow. The density value is normalized for each main scanning line for the histogram having a narrow density range (step S701). Since normalization can be performed by searching the density value information of the pixel of interest, it is performed for each pixel. The normalization formula is as follows: Normalized density value = (maximum gradation value of input image- (brightness value before normalization x maximum gradation value of input image) / maximum brightness value of input image) , Widen the difference in density between the text area and background area. After that, the density histogram is created (step S103), the density value having the highest appearance frequency for each line is searched (step S104), and the background is removed by the density conversion formula for one line (step S105). Perform background removal processing.

FIG. 8 shows a flow of background removal processing according to claim 4 performed by the signal processing unit 903 of FIG. With respect to the multi-valued document image 101 input in the processing order of FIG. 8, a density histogram is created for each line in the main scanning direction (step S10).
3) is performed, and the density value having the maximum appearance frequency for each line is searched from this density histogram (step S104).
Based on the density information of the maximum frequency for each line, the background distribution is removed from the pixels distributed in that line by the density conversion formula of one line (step S105). For the pixels distributed in that line, the density conversion formula is changed to a or b depending on the density value of the pixel of interest to remove the background. The above-described processing is performed in all main scanning directions to output the document image 102 from which the background is removed.

First, the multi-valued document image 101 is searched for a density value for one line in the main scanning direction, and a density histogram is created (step S103) to create a density histogram for one line. Thereby, the density distribution information of one line in the main scanning direction is stored. This is used as information of density conversion for pixel regions distributed in one main scanning line.

Next, the density histogram of one line in the main scanning is searched for the density value having the maximum appearance frequency for each line (step S104). The density histogram of one line in the main scan obtained in the creation of the density histogram (step S103) is searched for, and the density value with the maximum appearance frequency of one line is stored. This value is used in the density conversion formula of the next process. When the density value having the highest appearance frequency of one line in the main scanning direction is stored, the density conversion formula is changed for each density value of the pixel of interest, and the background removal by the density value conversion formula (step S8).
01, S802) to remove the background. Since the document image is more distributed in the background area, the density value with the highest appearance frequency for each line is assumed to be the background density value of that line, and this value is used as a threshold to determine whether the density value is low. The density value conversion equation a for the area and the density value conversion equation b for the background area when the density value is high are separately processed. 1
The density value with the maximum line appearance frequency and the minimum density value in the entire image in all main scanning directions are searched before this processing,
Using these two values, the density value of the pixel of interest of the input image is used in the conversion formula for conversion. For pixels belonging to one main scanning direction, the density value with the highest appearance frequency of the line is used, and the density value with the highest appearance frequency for each line in all the main scanning directions and the minimum density value for the entire image are used. The density value is converted for one line to remove the background. When the density value conversion process is completed for each line, the density histogram for each line is created again for the next line (step S103), and the density value with the highest appearance frequency for each line is searched (step S10).
4) The background removal is performed by the density conversion formula for one line (step S105). When all line processing is completed, the background removal processing is completed and the processed image is output.
Pixels in the character area have a low density value before processing, and since density conversion is performed using the density conversion equation a for the character area, the density value changes little, but the pixels in the background area do Since it is converted by, the density value becomes lower than the density value before conversion.

According to the above-described embodiment, the background removal algorithm of the present invention uses the density value of the maximum frequency of one line on the vertical axis main scanning and the histogram of the number of lines on the horizontal axis sub-scanning as the feature amount of background removal, The density conversion formula is used for each scanning line to remove the background according to the feature amount. Furthermore, it is possible to process the information of the entire image line by line without scanning once as in the conventional method. Also, even if the background density is not uniform, only the local information is used, so it is not affected by the overall density information, and the background can be removed in a local area. Is. Even when reading character information from a document image having a high background density value, an image having a high difference in density value between the character area and the background area is clearer and easier to read.

[0033]

According to the invention described in claim 1, since the density value of the background of the input image can be specified by using the density histogram having the maximum appearance frequency for each line, the background value of the background can be specified based on the information. The density value can be converted to a low value. Furthermore, since it is possible to perform processing line by line, the amount of memory used for image processing is reduced.

According to the second aspect of the present invention, the density histogram having the maximum appearance frequency for each line is smoothed to smooth the difference between the neighboring areas, so that the background of the output image is removed. In the image, the difference between the density values in the vicinity is small, and an unnatural edge is less likely to occur. Further, the density conversion formula lowers the density value of the background and removes the background.
Furthermore, since it is possible to perform processing every few lines, the amount of memory used for image processing is reduced.

According to the third aspect of the present invention, by first normalizing the minimum density value of the input image to the minimum density value of the gradation, the minimum density value approaches the origin, so that an image with a high background density value. Since the background density value can be effectively set to a low level, the background of the output image is removed. Furthermore, since it is possible to perform processing line by line, the amount of memory used for image processing is reduced.

According to the fourth aspect of the present invention, the background density value is lowered by changing the density conversion equation using the density histogram having the maximum appearance frequency for each line and the information of the density value of the input image. Since the density of the character changes little and the density can be converted, it is possible to output the image in which the background of the output image is removed and the density of the character is the same as the input image. Furthermore, since it is possible to perform processing line by line, the amount of memory used for image processing is reduced.

[Brief description of drawings]

FIG. 1 is a flowchart showing a background removal process according to claim 1 of the present invention.

FIG. 2 is a diagram showing a document image before background removal according to the present invention and an image after background removal.

FIG. 3 is a graph showing a density histogram of the images of FIGS. 3 (a) and 3 (b) according to the present invention.

FIG. 4 is a flowchart showing a background removal process according to claim 2 of the present invention.

FIG. 5 is a diagram showing a part of the density values having the maximum appearance frequency for each line according to the present invention.

FIG. 6 is a diagram showing comparison of the presence or absence of smoothing processing of the present invention.

FIG. 7 is a flowchart showing a background removal process according to claim 3 of the present invention.

FIG. 8 is a flowchart showing a background removal process according to claim 4 of the present invention.

FIG. 9 is a diagram showing a configuration of a background removal processing device.

[Explanation of symbols]

101 Multi-valued document image 102 Document image with background removed 103 Creating a density histogram for each line 104 Density value with the highest appearance frequency per line 105 Concentration conversion formula for each one-line density conversion

Claims (4)

[Claims] 1. An image input section for digitizing and capturing character information, and an image signal processing section for performing image processing calculation,
In a system having a storage unit for storing processing results,
A means for creating a density histogram for each line in the main scanning direction, a means for searching for a density value having a maximum appearance frequency for each line in the main scanning direction, and an appearance frequency for each line in the main scanning direction An image processing apparatus comprising: means for converting a density value of an image based on maximum density value information. 2. The density value having the maximum appearance frequency for several lines obtained by the means for obtaining the density value having the maximum appearance frequency for each line in the main scanning direction,
A means for performing smoothing is provided.
The image processing device described. 3. The image processing apparatus according to claim 1, further comprising a unit for normalizing the density value before the unit for obtaining the density value having the maximum appearance frequency for each line in the main scanning direction. . 4. A means for converting a density value of an image based on density value information having the maximum appearance frequency for each line in the main scanning direction, and a means for changing the means for converting the density value using the density information. The image processing apparatus according to claim 1, further comprising: