JP7357853B2 - Image processing device, image processing method, and image processing program - Google Patents

Description

The present invention relates to an image processing device, an image processing method, and an image processing program, and particularly relates to a binarization processing technique.

When a document containing background noise is read and the read image is subjected to binarization processing, there is a problem that unnecessary background noise remains. To solve this problem, for example, Patent Document 1 calculates a moving average of the frequency for each density in the density histogram of an input image, calculates a binarized slice level based on the resulting density value for the maximum frequency, and We are proposing a technology for converting values into values. However, the threshold is set using a simple method on the assumption that the histogram after taking the moving average has clear bimodality, so if there is no clear bimodality, the threshold cannot be set. The problem is that I can't. On the other hand, in Patent Document 2, a "cell" of n×n pixels is acquired, a measurement area of m×m pixels (n<m) is set around the cell, and a local area for the cell is determined from a brightness histogram in the measurement area. proposed a technique to obtain a local threshold value and binarize cells using a local threshold value within a preset limit value. The reason for setting the value within the preset limit value is to suppress extraction of noise components.

Japanese Patent Application Publication No. 7-262348 Japanese Patent Application Publication No. 2001-245148

However, according to the knowledge of the inventor of the present application, when the density distribution of the background is wide, clear bimodality is often not recognized, and Patent Document 1 deals with the binarization of such an image. The problem is that I can't. On the other hand, according to the inventor's analysis, since the generation of noise components varies depending on the density of the background, it is difficult to extract noise components and eliminate thin figures and characters using fixed limit values set in advance. The problem is that we cannot escape the problem of trade-offs.

The present invention has been made in view of this situation, and an object of the present invention is to provide a binarization processing technique that improves both the discrimination of low-density images and the suppression of image noise.

The image processing device of the present invention uses a discriminant analysis method for input image data to determine a first average gradation value, which is an average gradation value of a first class, and a first average gradation value, which is an average gradation value of a first class, and a first average gradation value, which is an average gradation value of a first class, and a first average gradation value, which is an average gradation value of a first class, and a first average gradation value, which is an average gradation value of a first class, for input image data. a histogram analysis unit that calculates a second average gradation value that is an average gradation value of the second class; and a threshold value is adjusted between the first average gradation value and the second average gradation value. an operation display unit that accepts user input to perform the operation; and binarization that adjusts the threshold according to the user input input to the operation display unit and executes binarization processing using the adjusted threshold. and a processing section, and the operation display section displays the image after the binarization process in real time according to the operation of the threshold value.

The image forming method of the present invention uses a discriminant analysis method for input image data to determine a first average gradation value, which is an average gradation value of a first class, and a first average gradation value, which is the average gradation value of the first class, and a first average gradation value, which is the average gradation value of the first class, and a first average gradation value, which is the average gradation value of the first class, and a first average gradation value, which is the average gradation value of the first class, and a histogram analysis step of calculating a second average gradation value that is an average gradation value of the second class; and adjusting a threshold between the first average gradation value and the second average gradation value. an operation/display step for accepting user input for the operation and display; and a binarization step for adjusting the threshold according to the user input input in the operation/display step, and performing a binarization process using the adjusted threshold. and a processing step, and the operation display step displays the binarized image in real time according to the operation of the threshold value.

The image processing program of the present invention uses a discriminant analysis method for input image data to determine a first average gradation value, which is an average gradation value of a first class, and a first average gradation value, which is an average gradation value of a first class, and a first average gradation value, which is an average gradation value of a first class. a histogram analysis unit that calculates a second average gradation value that is an average gradation value of the second class, and adjusts a threshold between the first average gradation value and the second average gradation value; an operation display unit that accepts user input for the operation, and a binarization process that adjusts the threshold according to the user input input to the operation display unit and executes the binarization process using the adjusted threshold. The operation display section displays the binarized image in real time according to the operation of the threshold value.

According to the present invention, it is possible to provide a binarization processing technique that improves both the discrimination of low-density images and the suppression of image noise.

1 is a schematic configuration diagram showing the overall configuration of an image forming apparatus 10 according to a first embodiment of the present invention. 1 is a block diagram showing a functional configuration of an image forming apparatus 10 according to a first embodiment. FIG. 3 is an explanatory diagram showing a read image and a brightness histogram of the read image. FIG. 7 is an explanatory diagram showing a binarization process according to a comparative example and a threshold value THc for the binarization process of the comparative example. It is a flowchart which shows the content of the binarization processing procedure based on 1st Embodiment. FIG. 3 is an explanatory diagram showing the contents of low-pass filter processing according to the first embodiment. It is a flowchart which shows the content of page threshold value setting processing concerning a 1st embodiment. It is an explanatory diagram showing the contents of page threshold value setting processing concerning a 1st embodiment. FIG. 7 is an explanatory diagram illustrating the content of background noise determination threshold setting processing according to the first embodiment and the comparative example. FIG. 3 is an explanatory diagram showing the content of background noise removal processing according to the first embodiment and the comparative example. FIG. 3 is an explanatory diagram showing the contents of background noise removal processing and binarization processing according to the first embodiment. It is a flowchart which shows the content of the binarization process based on 1st Embodiment. FIG. 3 is an explanatory diagram showing how to select a local region of interest according to the first embodiment. It is a graph which shows the threshold value THS for switching based on 1st Embodiment. It is an explanatory diagram showing the contents of binarization processing concerning a 1st embodiment. It is a flowchart which shows the content of the threshold value adjustment process for binarization based on 2nd Embodiment. FIG. 7 is an explanatory diagram showing an operation display screen according to the second embodiment. It is an explanatory view showing the contents of filter processing concerning a modification.

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, modes for carrying out the present invention (hereinafter referred to as "embodiments") will be described in the following order with reference to the drawings.
A. First embodiment:
B. Second embodiment:
C. Variant:

A. First embodiment:
FIG. 1 is a schematic configuration diagram showing the overall configuration of an image forming apparatus 10 according to a first embodiment of the present invention. FIG. 2 is a block diagram showing the functional configuration of the image forming apparatus 10 according to the first embodiment. The image forming apparatus 10 includes an image reading section 100, a control section 210, an image forming section 220, an operation display section 230, a storage section 240, and a FAX processing section 250. The image reading unit 100 includes an automatic document feeder (ADF) 160 and a document table (contact glass) 150, reads an image (original image) from a document, and generates an image data ID that is digital data.

The image forming unit 220 forms an image on a print medium (not shown) based on the image data ID and discharges the print medium. The image forming section 220 includes a color conversion processing section 221, a halftone processing section 222, and an image output section 223. The color conversion processing unit 221 converts the image data ID, which is RGB data, into CMYK image data. The halftone processing unit 222 performs halftone processing to generate halftone data of CMYK image data. The image output unit 223 forms an image based on the halftone data. The operation display unit 230 receives user operation inputs from a display (not shown) functioning as a touch panel and various buttons and switches (not shown).

The control unit 210 includes main storage means such as RAM and ROM, and control means such as MPU (Micro Processing Unit) and CPU (Central Processing Unit). Further, the control unit 210 has controller functions related to interfaces such as various I/Os, a USB (universal serial bus), a bus, and other hardware, and controls the entire image forming apparatus 10 .

The storage unit 240 is a storage device that is a non-temporary recording medium such as a hard disk drive or flash memory, and stores control programs (including image processing programs) for processing executed by the control unit 210 and the binarization processing unit 140. and memorize data.

The image reading unit 100 includes a light source driver 111 and a light source 112, as shown in FIG. The light source 112 includes a plurality of LEDs (not shown) that irradiate the document D with light. The light source driver 111 is an LED driver that drives a plurality of LEDs arranged in the main scanning direction, and performs on/off drive control of the light source 112. Thereby, the light source 112 can irradiate the document surface of the document D with the irradiation light L1 using pulse width modulation (PWM) with variable drive duty.

The irradiation light L1 is irradiated at an angle of 45 degrees (in a tilted direction) with respect to a direction perpendicular to the surface of the document D. The document D reflects reflected light including diffusely reflected light L2 and specularly reflected light. The light receiving element 122 receives the diffusely reflected light L2.

As shown in FIG. 1, the image reading unit 100 further includes a first reflecting mirror 113, a first carriage 114, a second reflecting mirror 115, and a third reflecting mirror between the document D and the image sensor 121. It includes a mirror 116, a second carriage 117, and a condenser lens 118. The first reflecting mirror 113 reflects the diffusely reflected light L2 from the document D toward the second reflecting mirror 115. The second reflecting mirror 115 reflects the diffusely reflected light L2 in the direction of the third reflecting mirror 116. The third reflecting mirror 116 reflects the diffusely reflected light L2 in the direction of the condenser lens 118. The condensing lens 118 forms an image of the diffusely reflected light L2 on each light receiving surface (not shown) of a plurality of light receiving elements 122 included in the image sensor 121.

The image sensor 121 is three CCD line sensors (not shown) that detect each of the three colors R, G, and B using color filters (not shown) for each color component of R, G, and B. . The image sensor 121 scans (sub-scans) a document with three CCD line sensors extending in the main scanning direction and acquires an image on the document as a combination of voltage values corresponding to R, G, and B. In this way, the image sensor 121 can perform photoelectric conversion processing and output R, G, and B analog electrical signals for each pixel in the main scanning direction.

The first carriage 114 carries a light source 112 and a first reflecting mirror 113, and reciprocates in the sub-scanning direction. The second carriage 117 carries a second reflecting mirror 115 and a third reflecting mirror 116, and reciprocates in the sub-scanning direction. The first carriage 114 and the second carriage 117 are controlled by a control section 210 that functions as a scanning control section. Thereby, the light source 112 can scan the original in the sub-scanning direction, so the image sensor 121 can output an analog electrical signal in accordance with the two-dimensional image on the original.

Note that when the automatic document feeder (ADF) 160 is used, the first carriage 114 and the second carriage 117 are fixed at a preset sub-scanning position, and the document D is automatically fed in the sub-scanning direction. is scanned. Note that some ADFs 160 read not only one side but also both sides simultaneously or sequentially.

The ADF 160 includes a paper feed roller 161 and a document reading slit 162. The paper feed roller 161 automatically feeds the original, and the original is read through the original reading slit 162. In this case, since the first carriage 114 is fixed at a preset sub-scanning position, the light source 112 mounted on the first carriage 114 is also fixed at a predetermined position.

As shown in FIG. 2, the image reading section 100 further includes a signal processing section 123, a shading correction section 124, a shading correction table 124a, a gamma conversion section 125, a binarization processing section 140, and an AGC processing section. 130, and a white reference plate 132 (see FIG. 1). The binarization processing section 140 includes a filter processing section 141 and a threshold value setting section 142.

The signal processing section 123 is a variable gain amplifier having an A/D conversion function. The signal processing section 123 amplifies the analog electrical signal with the gain set by the AGC processing section 130 and stored in the storage section 240, and A/D converts the amplified analog electrical signal into digital data.

In this embodiment, the AGC processing unit 130 is a gain adjustment unit that sets optimal gains and offset values for each of the plurality of light receiving elements 122 using a black reference signal and a white reference signal. The black reference signal is an analog electrical signal of the light receiving element 122 when the light source 112 is off. The white reference signal is an analog electrical signal of the light receiving element 122 when the white reference plate 132 is irradiated instead of the original D.

The AGC processing unit 130 sets the offset value so that each RGB gradation value of the image data ID becomes the minimum value “0” when the black reference signal is A/D converted by the signal processing unit 123. The AGC processing unit 130 uses this offset value to set the gain so that each RGB gradation value of the image data ID becomes the maximum value “255” when the white reference signal is A/D converted by the signal processing unit 123. Set. Thereby, the dynamic range from the minimum value "0" to the maximum value "255" can be effectively utilized.

The shading correction unit 124 performs shading correction on the digital data to generate image data ID. Shading correction is performed to correct shading that occurs due to non-uniformity of light amount in the length direction of the light source 112, marginal light reduction due to the cosine fourth power law of the lens, and uneven sensitivity of the plurality of light receiving elements 122 arranged in the main scanning direction. This is a correction to suppress the The gamma conversion unit 125 performs gamma conversion based on the characteristics of the image reading unit 100. Thereby, the image reading unit 100 can generate image data ID having each RGB tone value.

The binarization processing unit 140 uses the calculation formula for the brightness value L (brightness value L≈0.3R+0.6G+0.1B) and calculates the brightness value L using each RGB gradation value of the image data ID. Then, monochrome image data MD0 is generated. The binarization processing unit 140 further performs binarization processing on the monochrome image data MD0 to generate binary data BD in which each pixel is represented by 1 bit. The FAX processing unit 250 can perform facsimile transmission processing using binary data BD having a small data size.

In this way, the image reading unit 100 reads the image on the document D, generates the image data ID, and generates the binary data BD as necessary. The image data ID is RGB image data that represents the image on the document D using RGB gradation values (0 to 255).

FIG. 3 is an explanatory diagram showing a read image and a brightness histogram of the read image. FIG. 3(a) shows the original image D0. The original image D0 is an image with relatively low background brightness and poor visibility. The original image D0 is an image represented by image data ID generated by reading by the image reading section 100.

FIG. 3(b) shows a brightness histogram H0 of the monochrome image data MD0. Monochrome image data MD0 is a histogram of brightness values L of image data ID representing original image D0. The horizontal axis of the brightness histogram is the brightness gradation value (0 (black) to 255 (white)), and the vertical axis is the number of pixels (frequency). The brightness histogram H0 includes a line pixel group that is a pixel group that makes up a line drawing image such as black text or a ruled line, and a background pixel group that is a pixel group that makes up a background image.

FIG. 4 is an explanatory diagram showing the binarization process according to the comparative example and the binarization process threshold THc of the comparative example. The binarization processing threshold THc is set using a discriminant analysis method (Otsu's method) as a comparative example. Discriminant analysis is a method that divides a brightness histogram into two classes using a provisional threshold value, examines the shape of the histogram from the variance values within each class and the variance values between classes, and adopts the threshold value that is thought to be located in the lowest valley. be. Note that the processing details of the discriminant analysis method will be described later.

FIG. 4A shows a binarized image Dc as a result of the binarization process according to the comparative example. The binarized image Dc is an image represented by the binary data BD after the binarization process. The binary data BD is processed by the binarization processing unit 140 to set the brightness gradation value of the line pixel group of the monochrome image data MD0 as input image data to "0 (black)", and to perform the binarization processing on the background pixel group. Image data that is generated by setting the luminance gradation value of a pixel whose luminance gradation value is lower than the threshold value THc to "0", while setting the luminance gradation value of other pixels to "1 (white)". be.

FIG. 4(b) shows the binarization processing threshold THc according to the comparative example. The binarization processing threshold THc not only discriminates a black text image as a pattern area, but also discriminates an image area composed of pixels with low brightness in a background image as a pattern area. The brightness of the background image determined to be a pattern area is changed to "0 (black)" by the binarization processing unit 140, so that the binarized image Dc has noticeable graininess due to the binarization process. The image is deteriorated (rough) (see FIG. 4(a)).

FIG. 5 is a flowchart showing the contents of the binarization processing procedure according to the first embodiment. The binarization process according to the first embodiment can significantly suppress deterioration of graininess caused by the background image as described above.

In step S100, the filter processing unit 141 of the binarization processing unit 140 performs low-pass filter processing. In the low-pass filter processing, the binarization processing unit 140 performs smoothing processing on, for example, 300 dpi monochrome image data MD0 (also referred to as input image data) using a 5×5 pixel Gaussian filter. Thereby, the low-pass filter processing can smooth noise in the background image and generate monochrome image data MD1. Noise in the background image is an example of background noise.

FIG. 6 is an explanatory diagram showing the contents of the low-pass filter processing according to the first embodiment. FIG. 6A shows a brightness histogram H0 of monochrome image data MD0 before low-pass filter processing. FIG. 6(b) shows a brightness histogram H1 of the monochrome image data MD1 after low-pass filter processing. As can be seen from the brightness histogram H1, the gradation value distribution of the discrete background pixel group is gathered into a single peak.

The low-pass filter process smoothes the high frequency component (the gradation area of the background where the pixel value fluctuates significantly (the noisy and jagged brightness component)) of the monochrome image data MD0, and thereby performs the background noise determination threshold setting process (step S300). ) to be able to set appropriate thresholds.

In step S200, the threshold value setting unit 142 of the binarization processing unit 140 sets the page threshold value THP based on the brightness histogram H1 after the low-pass filter processing. In the page threshold setting process, the binarization processing unit 140 sets a page threshold using a discriminant analysis method (Otsu's method). The page threshold THP is a binarization threshold that can be applied to the entire image.

FIG. 7 is a flowchart showing the contents of the page threshold setting process (step S200) according to the first embodiment. FIG. 8 is an explanatory diagram showing the contents of the page threshold setting process according to the first embodiment. In step S210, the binarization processing unit 140 functions as a histogram analysis unit, and uses the provisional threshold to temporarily divide the brightness histogram H1 into class 1 (also called the first class) and class 2 (second class). (also called class) (see FIG. 8(a)). The second class is a class that includes a relatively large number of background images.

In step S220, the binarization processing unit 140 executes intra-class variance calculation processing. In the intra-class variance calculation process, the binarization processing unit 140 calculates the intra-class variance using calculation formula F1. In step S230, the binarization processing unit 140 executes an inter-class variance calculation process. In the inter-class variance calculation process, the binarization processing unit 140 calculates the inter-class variance using calculation formula F2. In step S240, the binarization processing unit 140 executes total variance calculation processing. In the total variance calculation process, the binarization processing unit 140 calculates the total variance using calculation formula F3. The total variance can be used as an evaluation value for the provisional threshold.

The binarization processing unit 140 repeatedly executes the processing of steps S210 to S240 for all the luminance gradation values (step S250). In step S260, the binarization processing unit 140 sets the page threshold THP to the gradation value with the maximum total variance.

In step S300, the binarization processing unit 140 functions as a background noise determination threshold setting unit and sets a background noise determination threshold THB, which is a threshold for background noise determination. Note that this process (step S300) may be skipped when the background is black. The background noise determination threshold THB is also called a background noise determination threshold.

FIG. 9 is an explanatory diagram showing the contents of the background noise determination threshold setting process according to the first embodiment and the comparative example. FIG. 10 is an explanatory diagram showing the details of the background noise removal processing according to the first embodiment and the comparative example. FIG. 9A shows a method for setting the background noise determination threshold THB according to the first embodiment. FIG. 9B is an explanatory diagram showing the contents of the background noise determination threshold setting process according to the comparative example. FIG. 10A is an explanatory diagram showing the details of the background noise removal process according to the first embodiment. FIG. 10(b) is an explanatory diagram showing the details of the background noise removal process according to the comparative example.

In this example, the binarization processing unit 140 generates a brightness gradation value that is the number of pixels (frequency) of a predetermined range (for example, 10%) of the peak pixel number, which is the number of pixels of the peak value PB of the distribution of the background image. is searched from the side of the background peak gradation value, which is the luminance gradation value of the peak value PB, to the side of the lower luminance gradation value (the side where the density becomes higher).

In the background noise determination threshold THB (FIG. 9(b)) according to the comparative example, the gradation value distribution of the background pixel group is discrete (jagged), so the binarization processing unit 140 eliminates the background noise. It can be seen that it is not possible to set a threshold for background noise determination that can effectively eliminate background noise (see FIG. 10(b)). On the other hand, in the background noise determination threshold THB (FIG. 9(a)) according to the first embodiment, the gradation value distribution of the background pixel group is gathered into a single peak, so the binarization processing unit 140 can effectively eliminate background noise by smoothly setting an appropriate background noise determination threshold (FIG. 10(a)).

Thereby, the binarization processing unit 140 can set the searched threshold value as the background noise determination threshold value THB. In the brightness histogram H1, as described above, the high frequency component (the gradation region of the background where pixel values fluctuate significantly (the noisy and jagged brightness component)) is smoothed, so the binarization processing unit 140 It is possible to smoothly set the background noise determination threshold THB so that the residual amount does not become excessive. The ratio in the predetermined range is preferably 5% to 20%, more preferably 5% to 15%, and most preferably around 10%.

In step S400, the binarization processing unit 140 compares the page threshold THP and the background noise determination threshold THB, and if the page threshold THP is larger than the background noise determination threshold THB, the process proceeds to step S500. , if the page threshold THP is less than or equal to the background noise determination threshold THB, the process advances to step S600.

In step S500, the binarization processing unit 140 replaces and resets the value of the background noise determination threshold THB with the value of the page threshold THP. As a result, the binarization processing unit 140 can detect that when the background noise determination threshold THB is set too small, the background noise is excessively erased in the background noise removal process, and text images etc. that should be left are deleted. You can prevent what happened.

In step S600, the binarization processing unit 140 functions as a background noise reduction processing unit and executes background noise removal processing (also referred to as background noise reduction processing). In the background noise removal process, the binarization processing unit 140 converts the pixel value of a pixel whose luminance gradation value is higher (lower density) than the background noise determination threshold THB into the luminance gradation value of the background noise determination threshold THB. Set.

Thereby, the binarization processing unit 140 can generate monochrome image data MD2 having the brightness histogram H2. In the brightness histogram H2, the brightness gradation value distribution of the background image is smoothed to concentrate on the background noise determination threshold THB, and the contrast becomes low. Misidentification with pattern images can be suppressed.

FIG. 11 is an explanatory diagram showing the contents of the background noise removal process and the binarization process according to the first embodiment. In this example, unlike the binarization process (step S700) described later, it is assumed that the binarization processing unit 140 performs processing by fixing the threshold value of the binarization process to the page threshold value THP. FIG. 11A shows a monochrome image D1 after background noise removal processing. The monochrome image D1 is processed so that the luminance of pixels with significantly low luminance increases while the luminance of pixels with significantly high luminance decreases in the background image, so the graininess of the monochrome image has improved (fine ) has a background image.

FIG. 11(b) shows the binary image D2 after the binarization process according to the first embodiment. By setting the page threshold value THP according to the first embodiment, the binary image D2 has a significantly reduced number of pixels in the background area that are misidentified as pixels in the pattern area, so that the binary image D2 becomes an appropriate image without deteriorating graininess. It has become. Such a reduction in misjudgment is achieved by reducing graininess and reducing local contrast through background noise removal processing.

In this way, the image forming apparatus 10 according to the first embodiment executes low-pass filter processing as pre-processing before setting the threshold value for binarization processing. Thereby, the binarization processing unit 140 can reduce the noise in the background image and reduce the variance of the brightness gradation values. Removal of noise from the background image can reduce deterioration in graininess caused by binarization processing. Reducing the variance of the brightness gradation values is achieved by narrowing the width of the distribution of the background image around the average value of the brightness gradation values of the background image, and by smoothing the changes in the gradation values of the distribution to set the background noise determination threshold THB. settings can be made more smoothly.

FIG. 12 is a flowchart showing the contents of the binarization process (step S700) according to the first embodiment. In step S710, the binarization processing unit 140 functions as a local region of interest selection unit and converts a plurality of pixels constituting the monochrome image data MD2 (also referred to as input image data) into (M×N) pixels (M and N (for example, an integer of 3 or more), and the divided local regions are sequentially selected as local regions of interest. The local area of interest is, for example, an area with a reference size of 5×5 pixels. The reference size is set depending on, for example, the resolution of the image to be processed.

FIG. 13 is an explanatory diagram showing how to select a local region of interest according to the first embodiment. FIG. 14 is a graph showing the switching threshold THS according to the first embodiment. In this example, two local regions LR1 and LR2 are shown. When the local region LR1 is selected as the local region of interest, the black text image is within the reference range, so the minimum brightness value is close to 0, and the maximum brightness value is the brightness value of the background, that is, the background noise determination. The brightness value is near the threshold value THB. On the other hand, when the local region LR2 is selected, since the thin text image is within the reference range, the minimum brightness value becomes the brightness value of the thin text image, and becomes the brightness value near the background noise determination threshold THB, The maximum brightness value is the brightness value of the background.

In step S720, the binarization processing unit 140 executes maximum brightness value acquisition processing. In the maximum brightness value acquisition process, the binarization processing unit 140 acquires the maximum brightness value within the local area of interest. Specifically, the binarization processing unit 140 obtains the maximum brightness value of the background in each local area, regardless of whether the local area of interest is the local area LR1 or the local area LR2. In this example, it is assumed that the binarization processing unit 140 has acquired the local maximum brightness value L_Max (brightness value 170) in the local region LR2 (see FIG. 14(a)). The maximum luminance value is the maximum gradation value of luminance, and may be the maximum gradation value that is the density gradation value on the lowest density side.

In step S730, the binarization processing unit 140 functions as a switching threshold setting unit and executes switching threshold setting processing. In the switching threshold setting process, the binarization processing unit 140 reads out a preset binarization threshold switching threshold table (also referred to as a threshold switching table) T from the storage unit 240 and sets the switching threshold THS. Make it possible. In this example, it is assumed that the binarization processing unit 140 has acquired "30" as the switching threshold THS based on the local maximum luminance value L_Max (luminance value 170).

The threshold switching table T is a table obtained by discretizing the switching threshold THS defined by the calculation formula F4. The switching threshold THS indicates that the lower the brightness of the background area (the background image in this example), the more likely that the contrast occurring in the local area is not due to the pattern area but to noise in the image of the background area. This setting is based on the inventor's knowledge that the The inventor of the present application found through experiments and simulations that it is preferable to change the switching threshold THS according to the maximum gradation value. Specifically, it has been found that it is preferable to set the switching threshold THS to become larger as the maximum gradation value becomes smaller (as the density becomes darker), and particularly to set it so that it changes linearly.

In step S740, the binarization processing unit 140 executes local contrast calculation processing. In the local contrast calculation process, the binarization processing unit 140 calculates the difference between the maximum brightness value (170 in this example) and the minimum brightness value (120 in this example) of 25 pixels of 5×5 in the local area. is calculated as the local contrast LC (see calculation formula F5). In this example, it is assumed that the binarization processing unit 140 obtains "50" (=170-120) as the local contrast LC. The minimum luminance value is the minimum gradation value of luminance, and may be the minimum gradation value that is the gradation value on the darkest side.

In step S750, the binarization processing unit 140 advances the process to step S760 if the local contrast LC is greater than the switching threshold THS, and if the local contrast LC is less than or equal to the switching threshold THS, the binarization processing unit 140 continues the process. The process advances to step S770. Thereby, the binarization processing unit 140 can suppress erroneous determination that noise in the background area is the contrast of the pattern area and suppress hollow characters.

In step S760, the binarization processing unit 140 sets the local threshold THL (see calculation formula F6) as the binarization processing threshold used for the binarization processing of the local region LR2. In this example, in calculation formula F6, the coefficient α is, for example, a numerical value of 0.1 or more. In step S770, the binarization processing unit 140 sets the page threshold THP as the binarization processing threshold used for the binarization processing of the local region LR2. The processes from step S710 to step S770 are executed for all local regions (step S780). Note that the local threshold value may be set within a range between the maximum gradation value and the minimum gradation value.

FIG. 15 is an explanatory diagram showing the contents of the binarization process according to the first embodiment. FIG. 15(a) shows a document image D0a showing a thin text image. The original image D0a has the same background image as the original image D0 (see FIG. 3(a)), but the background image is shown with the background image removed to make the explanation easier to understand. FIG. 15(b) shows the image Dca after the binarization process according to the comparative example. In this example, in image Dca, a thin text image is erroneously determined to be a background image and disappears in the binarization process. FIG. 15(b) shows the image D2 after the binarization process according to the first embodiment. In this example, in image D2, a thin text image is determined to be an image of a pattern area and is clearly reproduced by binarization processing.

In this way, the image forming apparatus 10 according to the first embodiment adjusts the switching threshold THS according to the luminance gradation value according to the maximum luminance value in the local area, and locally adjusts the switching threshold THS based on the switching threshold THS. The binarization process is executed by appropriately switching between the threshold THL and the page threshold THP. Thereby, it is possible to improve both the discrimination of low-density images (such as thin text) and the suppression of image noise. On the other hand, since the present image forming apparatus 10 reduces the contrast of the background noise through the background noise removal process, it is also possible to suppress erroneous determination that noise in the background area is the contrast of a low-density image.

B. Second embodiment:
FIG. 16 is a flowchart showing the contents of the binarization threshold adjustment process (step S800) according to the second embodiment. In the binarization process according to the second embodiment, after the binarization process according to the first embodiment, the binarization threshold can be adjusted according to the user's wishes, and the binarization process can be executed again. It is configured as follows. The binarization process according to the second embodiment is different from the binarization process of the first embodiment in that a binarization threshold adjustment process (step S800) is added at the end of the process. It differs from the form.

FIG. 17 is an explanatory diagram showing an operation display screen according to the second embodiment. The binarization threshold adjustment process (step S800) is activated as an interactive fine adjustment process mode in response to a predetermined user input to the operation display unit 230 after the binarization process. The operation display section 230 has a user interface screen 231 and a start button 232. The user interface screen 231 has an interactive fine adjustment processing mode, and includes a selection icon 233, a cancel icon 234, an OK icon 235, a density decrease icon 237d, and a density increase icon 237u.

In step S810, the operation display unit 230 receives user input for region selection. The user can designate the selection region SR by touching and inverting the selection icon 233 and sliding his or her finger on the user interface screen 231. When reselecting the selection area, the designation of the selection area SR can be canceled by touching the cancellation icon 234. In this state, when the OK icon 235 is touched, the operation display section 230 displays a pop-up screen for character input (not shown).

In step S820, the operation display unit 230 accepts user input on a pop-up screen (not shown) for inputting characters. When the user inputs characters or touches an OK icon (not shown) on the pop-up screen without inputting characters, the process proceeds to step S830.

In step S830, the OCR processing unit 211 of the control unit 210 executes OCR accuracy calculation processing. In the OCR accuracy calculation process, the OCR processing unit 211 recognizes input characters, digitizes the OCR accuracy (recognition certainty), and calculates an integrated value of the recognition accuracy of each character. The OCR processing unit 211 sets a threshold value between the average brightness value M ₁ of class 1 (see FIG. 8) and the average brightness value M ₂ of class 2, which are set at the time of threshold TH1 and page threshold according to the first embodiment. The OCR accuracy is calculated for each threshold while changing the OCR accuracy (step S840). The OCR processing section 211 is also called a character recognition processing section. The average brightness value _M1 is also called the first average gradation value. The average brightness value _M2 is also called the second average gradation value.

In step S850, the OCR processing unit 211 of the control unit 210 sets the highest accuracy threshold. The highest accuracy threshold is a threshold with the highest cumulative value of recognition accuracy for each input character. The binarization processing unit 140 executes the binarization process on the selected region SR using the highest accuracy threshold value, and displays the binarized image after the binarization process on the user interface screen 231 in real time. Note that, for example, depending on the user's selection, the entire page may be subjected to the binarization process using the highest accuracy threshold.

In step S860, the operation display unit 230 receives user input for fine adjustment processing. In the fine adjustment process, touching the density decrease icon 237d allows fine adjustment to increase the threshold value, and touching the density increase icon 237u allows fine adjustment to decrease the threshold value.

In this manner, when the image forming apparatus 10 according to the second embodiment is not satisfied with the result of the automatic processing in the first embodiment, the average brightness value _M1 of class 1 and the class By automatically changing the threshold value with the highest OCR accuracy between the average brightness value M2 of 2 and the average brightness value _M2 of 2, it is possible to reproduce, for example, characters that disappeared during automatic binarization processing in the interactive fine-tuning processing mode. .

Thereby, the image forming apparatus 10 can easily deal with, for example, an abnormal image (all white or all black), and can also realize binarization processing according to the user's preference. Note that the density decrease icon 237d and the density increase icon 237u are displayed within the range of all gradation values or between the average brightness value _M1 and the average brightness value _M2 even if the processes of steps S810 to S850 are not executed. It is preferable to make it available.

C. Variant:
The present invention can be implemented not only in the above embodiment but also in the following modifications.

Modification 1: In the above embodiment, the filter used in the low-pass filter processing is a Gaussian filter, but is not limited thereto, and for example, a moving average filter may be used. Furthermore, if it has the characteristic of removing noise from a background image (for example, a background image), it is possible to remove noise and emphasize line drawings, such as with a DOG (Difference of Gaussian) filter (also called a Gaussian difference filter). Bandpass filters are also available.

FIG. 18 is an explanatory diagram showing the contents of filter processing according to a modified example. FIG. 18(a) is the same as FIG. 9(a), and shows the histogram H1 of the monochrome image data MD1 after Gaussian filter processing. FIG. 18(b) shows a histogram H1a of the monochrome image data MD1a after the DOG filter processing. In the histogram H1a, light text is emphasized. Thereby, in threshold setting for binarization processing, it is possible to set an appropriate threshold that increases the survivability of light text.

Modification 2: In the above embodiment, brightness gradation values and brightness histograms are used, but for example, density gradation values and density histograms may be used. The density is, for example, the density gradation value of the K color material when printing is performed using the K color material.

Modified Example 3: In the above embodiment, the present invention is applied to an image forming apparatus, but is not necessarily limited to an image forming apparatus, and is applicable to an electronic device that functions as an image processing apparatus such as an image reading apparatus or a mobile terminal. Applicable to equipment.

10 Image forming apparatus 210 Control section 220 Image forming section 230 Operation display section 240 Storage section 100 Image reading section 111 Light source driver 112 Light source 121 Image sensor 122 Light receiving element 123 Signal processing section 124 Shading correction section 130 AGC processing section 140 Binarization processing Section 141 Filter processing section 142 Threshold setting section 150 Document table 160 Automatic document feeder 162 Document reading slit

Claims (7)

a filter processing unit that performs low-pass filter processing as preprocessing on input image data before setting a threshold for binarization processing;
a threshold setting unit that sets a page threshold and a local threshold, which are set using a discriminant analysis method, as thresholds for binarization processing, based on the input image data that has been subjected to the low-pass filter processing;
A plurality of pixels included in the input image data subjected to the low-pass filter processing are divided into local regions composed of (M×N) pixels (M and N are integers of 3 or more), and the divided plurality of pixels are a local region of interest selection unit that sequentially selects local regions of interest as local regions of interest;
Obtain the maximum gradation value, which is the gradation value on the lowest density side among the gradation values of the plurality of pixels constituting the local area of interest, and switch so that the smaller the maximum gradation value is, the larger the gradation value becomes. a switching threshold setting unit that sets a threshold for
The local contrast, which is the difference between the maximum gradation value and the minimum gradation value among the gradation values of a plurality of pixels included in the local area of interest, is compared with the switching threshold, and the local contrast is determined by comparing the switching threshold with the local contrast. is larger than the switching threshold, a value obtained by multiplying the local contrast by a predetermined coefficient in the range between the maximum gradation value and the minimum gradation value is added to the minimum gradation value. Binarization processing that executes binarization processing using the set local threshold, and executes binarization processing using the page threshold when the local contrast is less than or equal to the switching threshold. Department and
A first average gradation value, which is the average gradation value of the first class, and a first average gradation value, which is the average gradation value of the first class, and a first average gradation value, which is the average gradation value of the first class, are determined by the discriminant analysis method on the input image data that has been subjected to the low-pass filter processing . a histogram analysis unit that calculates a second average gradation value that is an average gradation value of the second class;
an operation display section that accepts user input for specifying a selection area ;
a character recognition processing unit that recognizes characters included in the input image data in the selection area, calculates recognition accuracy that is the certainty of the recognition, and calculates an integrated value of the recognition accuracy of each character;
Equipped with
The binarization processing unit executes binarization processing using the changed threshold while changing the threshold between the first average gradation value and the second average gradation value . ,
The character recognition processing unit calculates the recognition accuracy for each of the changed thresholds, and sets a highest accuracy threshold that is a threshold with the highest cumulative value of recognition accuracy for each input character,
The binarization processing unit executes binarization processing of the input image data within the selected area using the highest accuracy threshold,
The operation display unit is an image processing device that displays the image after the binarization process. The image processing device according to claim 1, further comprising:
comprising a character recognition processing unit that recognizes characters included in the input image data, calculates recognition accuracy that is the certainty of the recognition, and calculates an integrated value of the recognition accuracy of each character,
The binarization processing unit changes the threshold between the first average gradation value and the second average gradation value,
The character recognition processing unit calculates the integrated value for each of the selected threshold values,
The binarization processing unit is an image processing device that executes binarization processing using a threshold value having the largest integrated value. The image processing device according to claim 1 or 2, further comprising :
A brightness histogram of the image data subjected to the low-pass filter processing, whose horizontal axis is the brightness gradation value and the vertical axis is the number of pixels, is generated, and the brightness histogram is calculated using the brightness gradation value of the peak value of the distribution of the background image of the brightness histogram. Background noise is determined by searching for a gradation value that has the number of pixels in a predetermined range of the number of pixels of the peak value, which is the number of pixels of the peak value, from the side of a certain background peak gradation value to the side of the gradation value where the density becomes darker. a background noise determination threshold setting unit that sets a threshold for
a background noise reduction processing unit that reduces background noise by setting all the gradation values of pixels having gradation values on the side where the density is lighter than the background noise determination threshold as the background noise determination threshold;
Equipped with
In the image processing apparatus , the low-pass filter processing is smoothing processing . The image processing device according to claim 3,
The filter processing unit is an image processing device that performs the low-pass filter processing using a Gaussian filter. The image processing device according to claim 3,
The filter processing unit is an image processing device that performs the low-pass filter processing using a Gaussian difference filter. a filter processing step of performing low-pass filter processing as preprocessing on input image data before setting a threshold for binarization processing;
a threshold setting step of setting a page threshold and a local threshold, which are set using a discriminant analysis method, as thresholds for binarization processing, based on the input image data on which the low-pass filter processing has been performed;
A plurality of pixels included in the input image data subjected to the low-pass filter processing are divided into local regions composed of (M×N) pixels (M and N are integers of 3 or more), and the divided plurality of pixels are a step of selecting a local region of interest as a local region of interest;
Obtain the maximum gradation value, which is the gradation value on the lowest density side among the gradation values of the plurality of pixels constituting the local area of interest, and switch so that the smaller the maximum gradation value is, the larger the gradation value becomes. a switching threshold setting step of setting a threshold for
The local contrast, which is the difference between the maximum gradation value and the minimum gradation value among the gradation values of a plurality of pixels included in the local area of interest, is compared with the switching threshold, and the local contrast is determined by comparing the switching threshold with the local contrast. is larger than the switching threshold, a value obtained by multiplying the local contrast by a predetermined coefficient in the range between the maximum gradation value and the minimum gradation value is added to the minimum gradation value. Binarization processing that executes binarization processing using the set local threshold, and executes binarization processing using the page threshold when the local contrast is less than or equal to the switching threshold. process and
A first average gradation value, which is the average gradation value of the first class, and a first average gradation value, which is the average gradation value of the first class, and a first average gradation value, which is the average gradation value of the first class, are determined by the discriminant analysis method on the input image data that has been subjected to the low-pass filter processing . a histogram analysis step of calculating a second average gradation value that is an average gradation value of the second class;
an operation display step of accepting user input for specifying a selection area ;
a character recognition processing step of recognizing characters included in the input image data in the selected area, calculating a recognition accuracy that is the certainty of the recognition, and calculating an integrated value of the recognition accuracy of each character;
Equipped with
In the binarization processing step, a threshold is changed between the first average gradation value and the second average gradation value, and the binarization processing is performed using the changed threshold . ,
The character recognition processing step calculates the recognition accuracy for each of the changed thresholds, and sets a highest accuracy threshold that is the threshold with the highest cumulative value of recognition accuracy for each input character,
The binarization processing step executes binarization processing of the input image data within the selected area using the highest accuracy threshold,
The operation display step is an image processing method of displaying the image after the binarization processing. a filter processing unit that performs low-pass filter processing as preprocessing on input image data before setting a threshold for binarization processing;
A threshold setting unit that sets a page threshold and a local threshold, which are set using a discriminant analysis method, as thresholds for binarization processing, based on the input image data on which the low-pass filter processing has been performed;
A plurality of pixels included in the input image data subjected to the low-pass filter processing are divided into local regions composed of (M×N) pixels (M and N are integers of 3 or more), and the divided plurality of pixels are a local region of interest selection unit that sequentially selects the local regions of as the local region of interest;
Obtain the maximum gradation value, which is the gradation value on the lowest density side among the gradation values of the plurality of pixels constituting the local area of interest, and switch so that the smaller the maximum gradation value is, the larger the gradation value becomes. a switching threshold setting unit that sets a threshold for
The local contrast, which is the difference between the maximum gradation value and the minimum gradation value among the gradation values of a plurality of pixels included in the local area of interest, is compared with the switching threshold, and the local contrast is determined by comparing the switching threshold with the local contrast. is larger than the switching threshold, a value obtained by multiplying the local contrast by a predetermined coefficient in the range between the maximum gradation value and the minimum gradation value is added to the minimum gradation value. Binarization processing that executes binarization processing using the set local threshold, and executes binarization processing using the page threshold when the local contrast is less than or equal to the switching threshold. Department,
A first average gradation value, which is the average gradation value of the first class, and a first average gradation value, which is the average gradation value of the first class, and a first average gradation value, which is the average gradation value of the first class, are determined by the discriminant analysis method on the input image data that has been subjected to the low-pass filter processing . a histogram analysis unit that calculates a second average gradation value that is an average gradation value of the second class;
an operation display unit that accepts user input for specifying a selection area ; and an operation display unit that recognizes characters included in the input image data within the selection area, calculates recognition accuracy that is the certainty of the recognition, and calculates the recognition accuracy of each character. causing the image processing device to function as a character recognition processing unit that calculates the cumulative value of the recognition accuracy;
The binarization processing unit executes binarization processing using the changed threshold while changing the threshold between the first average gradation value and the second average gradation value . ,
The character recognition processing unit calculates the recognition accuracy for each of the changed thresholds, and sets a highest accuracy threshold that is a threshold with the highest cumulative value of recognition accuracy for each input character,
The binarization processing unit executes binarization processing of the input image data within the selected area using the highest accuracy threshold,
The operation display section is an image processing program that displays the image after the binarization process.

Images