JP5979008B2 - Image processing apparatus, image processing method, and program - Google Patents

Description

  The present invention relates to an image processing apparatus, an image processing method, and a program for performing image processing.

  Conventionally, there is a technique for extracting a pattern from a captured image. For example, there are various techniques for extracting characters. Conventional character extraction methods are mainly classified into the following three types.

(1) Method based on edge extraction There is a technique in which a density image having only brightness is created from a color image, and a portion having a large change in brightness is extracted as a character line (for example, see Patent Document 1 and Non-Patent Document 1).

(2) There is a technique in which color clustering is applied to a method image based on color clustering to estimate and extract clusters corresponding to characters (see, for example, Patent Documents 2 to 3).

(3) Method based on linear approximation of color distribution Which color image has a high appearance frequency as the main color, connects the main colors with a line segment in the color space, and either of them by an internal dividing point obtained by projecting each pixel to the line segment There is a technique for extracting characters by determining whether they belong to the main color (see, for example, Patent Document 4).

JP 2000-181992 A JP 2009-199276 A JP 2009-191906 A JP 2003-16444 A

Bernsen, J, "Dynamic Thresholding of Gray-Level Images", Proc. Of the 8th Int. Conf. On Pattern Recognition, 1986

  Recently, there is a need to extract drug names from images in order to realize a drug management system in which drug packages are photographed with a digital camera, drug names are searched, and the effects and side effects are investigated.

  In many cases, a medicine package has a metal thin film affixed mainly on a plastic having lattice-like irregularities. Therefore, an image generated by imaging this package has a texture due to shadows and reflections even if the originally used colors are characters and background.

  The digital camera image is compressed in, for example, JPEG (Joint Photographic Experts Group) format, and color shift and noise occur.

  FIG. 1 is a diagram illustrating an example of an image compressed in the JPEG format. The example shown in FIG. 1 shows an image in which a predetermined material (for example, a material having irregularities on a metal foil and called material 1) on which the letter “A” is written is captured. Further, the predetermined material is light purple and the characters are red. As shown in FIG. 1, color shift and noise occur in an image in which light is reflected and an uneven material is captured.

  On the other hand, character extraction is performed on the image shown in FIG. 1 using conventional techniques (1) to (3). FIG. 2 is a diagram illustrating a result of character extraction using the method (1). As shown in FIG. 2, the texture portion of the background is extracted as a character, and the method (1) cannot appropriately extract a character from an image in which a character on the material 1 is captured.

  FIG. 3 is a diagram illustrating a result of character extraction using the method (2). As shown in FIG. 3, the extracted character pattern may be deficient or deformed due to the concentration of part of the character line due to JPEG compression or the hue of the character and the surroundings of the character being changed from the original hue. Yes. Therefore, in the method (2), it is not possible to appropriately extract characters from an image in which characters on the material 1 are captured.

  Next, a case where character extraction is performed on the material 1 using the method (3) will be described. Method (3) can deal with image degradation due to scanning and anti-aliasing, but cannot deal with textures caused by color shifts, shadows, and reflections.

  FIG. 4 is a diagram showing a distribution image in the RGB space of the image shown in FIG. When there is no texture or color shift caused by shadows or reflections, the distribution should be on an approximate line along the solid line arrow shown in FIG. However, the distribution is greatly deformed due to texture and color shift, and the distribution spreads in the direction along the dotted arrow. In this case, the distribution spreads particularly in the white direction.

  Therefore, in the method (3), it is not possible to appropriately extract characters from an image in which characters on the material 1 are captured.

  Therefore, when the conventional character extraction method is applied as it is to an image in which characters on material 1 are captured, there is a problem that character extraction cannot be performed appropriately.

  Accordingly, it is an object of the disclosed technique to provide an image processing apparatus, an image processing method, and a program that can appropriately perform character extraction.

  An image processing apparatus according to an aspect of the disclosure decomposes an image into a plurality of channels, separates the density distribution of each density image into two groups, a separation unit that generates a density image of each channel, and the degree of separation of the two groups A selection unit that selects a density image that maximizes the density, an extraction unit that extracts the maximum and minimum values of the density of the selected density image, and a graph cut that sets the extracted maximum and minimum values as a seed A separation unit that separates the selected density image into two groups, and an image generation unit that performs a connected component analysis on each of the two separated pixels and generates an image from any group of pixels; .

  According to the disclosed technique, character extraction can be performed appropriately.

The figure which shows an example of the image compressed in the JPEG format. The figure which shows the result of having performed the character extraction using the method (1). The figure which shows the result of having performed character extraction using the method (2). The figure which shows the distribution image in RGB space of the image shown in FIG. The figure which shows an example of the graph in the case of a 3x3 image. The figure which shows an example of the histogram of a foreground and a background. The block diagram which shows an example of the hardware of the image processing apparatus in an Example. The block diagram which shows an example of the function of the image processing apparatus in an Example. The figure which shows an example of a channel decomposition | disassembly result. The figure which shows the maximum separation degree in each channel. The figure which shows an example of the maximum value and the minimum value of the density | concentration image of the maximum resolution channel. The figure which shows an example of the lightness histogram and the density histogram of the separation degree maximum channel. The figure explaining the link and weight at the time of paying attention to the attention pixel. The figure which shows an example of the produced | generated binary image. 6 is a flowchart illustrating an example of image processing in the embodiment.

  Hereinafter, embodiments of an image processing apparatus, an image processing method, and a program will be described with reference to the accompanying drawings.

  First, the graph cut used in the embodiment will be described. The graph cut is a technique that generally cuts out an object, and is a technique that uses each pixel of an image as a node of the graph. For example, the Boycov et al. Method manually cuts an object based on a representative pixel of an object and a background (hereinafter also referred to as a seed) and a seed (for example, Y. Boycov and G. Funka-lea, “Interactive Graph Cuts for Optimal Boundary & Region Segmentation of Objects in ND Images, "Proc. Int'l Conf. Computer Vision, vol. L, pp. 105-112, July 2001.). Boycov's method has the disadvantage of having to manually set seeds.

  On the other hand, there is a technique of Han et al. That automatically calculates the seed (see Dongfeng Han, Wenhui Li, Xiaosuo Lu, Tianzhu Wang, and Yi Wang, "Automatic Segmentation Based on AdaBoost Learning and Graph-Cuts * ", ICIAR 2006, LNCS 4141, pp. 215-225, 2006).

  The Han et al. Method uses a different method to learn a target to be extracted in advance, roughly estimates the region, and then uses pixels inside and outside the region as seeds.

  However, when the object to be cut out is a pattern such as a character, there is a problem that an object whose type is enormous due to deformation due to character type or design is difficult to learn and is likely to be erroneously extracted or not extracted.

  In addition, since characters and patterns are line figures, it is difficult to correctly determine internal pixels. Therefore, Boycov's method and Han's method are considered inappropriate for the separation of characters and background, for example.

  Here, when a letter is printed on an uneven metal material like a medicine package, when this package is imaged, the background of the image is textured by reflections and shadows due to the unevenness, and the brightness A character pattern may not be distinguished by a change in value.

  Further, even when the color of the character line is not uniform due to deterioration due to JPEG compression, the character pattern may not be extracted with high accuracy.

  Therefore, if there is a character line in the brightness change part, the color of the character line is uniform, and the color of the pixel is distributed on the line segment connecting the representative color of the character and the background, a certain character for the grayscale image There are cases where the precondition of pattern extraction that there is always a threshold that can be extracted does not apply. Even when such a precondition is not satisfied, it is desirable that pattern extraction can be performed appropriately.

  Next, graph cut segmentation used in the embodiments described below will be briefly described. In the graph cut method, each pixel of an image is a node, and a graph having a node obtained by adding two nodes of a source node corresponding to a background (or foreground) label and a sink node corresponding to a foreground (or background) label. Created.

  A link called an N link is extended between adjacent pixel nodes, and a link called a T link is extended between the source node and the pixel node and between the sink node and the pixel node.

  FIG. 5 is a diagram illustrating an example of a graph in the case of a 3 × 3 image. The N link shown in FIG. 5 is given a smoothing cost that takes a value that decreases as the state between neighboring pixels becomes equal. Here, in the prior art, correct labels for the background and the foreground are assigned to a part of the pixel nodes manually or by learning. The correct answer label is also called a seed.

  A background seed is assigned a background correct label, and a foreground seed is assigned a foreground correct label. The aforementioned Boycov et al. Method is such that the user designates these seeds, and the Han et al. Method estimates the target object region by using these seeds by another method, and foreground seeds from the inside or outside of the region. Alternatively, the background seed is obtained.

  After obtaining the seed, the cost to be given to the T link is calculated from the relationship between the characteristics of the seed and the characteristics of the pixels other than the seed. The graph cut technique is used to separate the foreground and the background by cutting the graph into two subgraphs so that the energy function value defined by the sum of the costs of the T link and N link is minimized. is there.

  The energy function E is generally defined by the following equation (1).

Here, λ is a constant, R _i is a data cost given to the T link, B _ij is a smoothing cost given to the N link, and A is a label representing the background and foreground. D represents a pixel of the image, and N represents the vicinity of the pixel of interest.

After the seed is determined in some way, R _i is calculated as follows:
(I) The brightness of the pixel corresponding to each of the foreground seed and the background seed is obtained, and a brightness histogram is obtained. Here, the posterior probability P is obtained by dividing each frequency of the histogram by the total number of pixels of the corresponding seed.
(Ii) the pixel other than seed, when the brightness was I, using the histogram, the probability P _b to be I when the pixel was the background, the probability P _f which is I when was the foreground, respectively Ask.
(Iii) As shown by the following formulas (2) and (3), the data cost R _i between the source node and the sink node is calculated by taking the logarithm of each reciprocal of the probability value P. This data cost means that the higher the probability P, the smaller the cost.

FIG. 6 is a diagram illustrating an example of the foreground and background histograms. As shown in FIG. 6, if the probability P _b and the probability P _f at the lightness I ₁ are the same value, it is difficult to determine whether the pixel of the lightness I ₁ is assigned to the foreground or the background. .

Based on the above, when the above-described method using the graph cut is applied to, for example, character extraction, the following problems occur.
・ It is difficult for the user to set the seed each time by hand. ・ It is difficult to learn the shape of a character with thousands of character types and various fonts, and to accurately estimate the area (false extraction, unextracted). Extraction occurs)
・ Even if the text area can be estimated, it is difficult to determine the internal area that will be the foreground seed because it is a line figure. ・ When the seed can be estimated by some method, the lightness histograms of both the foreground and the background overlap. If the part is large, the accuracy is lowered. From the above, it is necessary to calculate the data cost without manually setting the seed or collecting and learning a large amount of data. Therefore, in the embodiment described below, the above-described graph cut method is used, but a method is realized in which a user can set a seed without manually setting a seed and without requiring learning.

[Example]
<Hardware>
FIG. 7 is a block diagram illustrating an example of hardware of the image processing apparatus 10 according to the embodiment. In the example illustrated in FIG. 7, the image processing apparatus 10 is connected to the imaging unit 20.

  The imaging unit 20 is, for example, a camera, and images characters printed on a medium in which a texture is generated in the background of the image due to reflections and shadows due to unevenness, for example. Note that the imaging unit 20 may capture an image on which a pattern is printed (or described).

  For example, the image processing apparatus 10 performs pattern extraction on an image captured by the imaging unit 20. The image processing apparatus 10 includes, for example, a control unit 101, a main storage unit 102, an auxiliary storage unit 103, a communication unit 104, and a drive device 105 as hardware. Each unit is connected via a bus so that data can be transmitted / received to / from each other.

  The control unit 101 is a CPU (Central Processing Unit) that performs control of each device, calculation of data, and processing in a computer. The control unit 101 is an arithmetic device that executes a program stored in the main storage unit 102 or the auxiliary storage unit 103. The control unit 101 receives data from the input device or the storage device, calculates, processes the output device, the storage device, and the like. Output to the device.

  The main storage unit 102 is, for example, a ROM (Read Only Memory) or a RAM (Random Access Memory). The main storage unit 102 is a storage device that stores or temporarily stores programs and data such as an OS and application software that are basic software executed by the control unit 101.

  The auxiliary storage unit 103 is, for example, an HDD (Hard Disk Drive) or the like, and is a storage device that stores data related to application software. The auxiliary storage unit 103 stores, for example, an image acquired from the imaging unit 20.

  The communication unit 104 performs data communication with peripheral devices by wire or wireless. The communication unit 104 acquires an image including a pattern via a network, for example, and stores it in the auxiliary storage unit 103.

  The drive device 105 reads a predetermined program from the recording medium 106, for example, a flexible disk or a CD (Compact Disc), and installs it in the storage device.

  Further, an image processing program to be described later is stored in the recording medium 106, and the program stored in the recording medium 106 is installed in the image processing apparatus 10 via the drive device 105. The installed image processing program can be executed by the image processing apparatus 10.

  In addition, although the image processing apparatus 10 has the imaging unit 20 as a separate configuration, it may be provided in the same apparatus, or may be provided with a display unit or the like.

<Function>
FIG. 8 is a block diagram illustrating an example of functions of the image processing apparatus 10 according to the embodiment. In the example illustrated in FIG. 8, the image processing apparatus 10 includes an image input unit 201, a decomposition unit 202, a selection unit 203, an extraction unit 204, a separation unit 205, and an image generation unit 206.

  Each unit illustrated in FIG. 8 functions when, for example, the control unit 101 loads an image processing program (extraction processing program) stored in the auxiliary storage unit 103 into the main storage unit 102 and executes it. That is, each unit illustrated in FIG. 8 can be realized by the control unit 101 and the main storage unit 102 as a work memory, for example. Further, each unit illustrated in FIG. 8 may be realized by hardware, for example. Hereinafter, character extraction will be described as an example of pattern extraction.

  The image input unit 201 inputs an image captured by the imaging unit 20, for example. The input image may be a color image. The image input unit 201 stores the input image in the main storage unit 102.

  The decomposition unit 202 decomposes the image input by the image input unit 201 into a plurality of channels, and generates a density image of each channel. The decomposition unit 202 may generate a plurality of density images by linear combination of image coordinate values. The decomposition unit 202 outputs each generated density image to the selection unit 203.

  The selection unit 203 separates the density distribution of each density image generated by the decomposition unit 202 into two groups, and selects a density image that maximizes the degree of separation of the two groups. The selection unit 203 includes a histogram calculation unit 231 and a channel selection unit 232 for performing this process.

  The histogram calculation unit 231 generates a density histogram from each density image of each channel. The histogram calculation unit 231 outputs the calculated density histograms to the channel selection unit 232.

  The channel selection unit 232, for each density histogram, for example, the degree of separation obtained by dividing the inter-group variance when the pixels are separated into two groups at each density value by the intra-group variance (separation degree S = inter-group variance / intra-group). Separation into two groups at a concentration that maximizes (dispersion). The channel selection unit 232 selects the density image of the channel having the largest degree of separation S among the largest degree of separation S of each density image. Since the channel and the density image have a one-to-one correspondence, the channel selection unit 232 may select a channel. The selection unit 203 outputs the selected density image to the extraction unit 204.

  The extraction unit 204 extracts the maximum value and the minimum value of the density of the density image selected by the selection unit 203. The extraction unit 204 outputs the extracted maximum value and minimum value to the separation unit 205.

  The separation unit 205 separates the selected density image into two groups using a graph cut in which the extracted maximum value and minimum value are set as seeds. The separation unit 205 includes a graph generation unit 251 and a graph cut unit 252.

  The graph generation unit 251 calculates and generates a graph in which the extracted maximum value and minimum value are set as seeds. The graph generation unit 251 outputs the calculated graph to the graph cut unit 252.

  The graph cut unit 252 cuts the graph into two subgraphs with respect to the graph generated by the graph generation unit 251 so that the sum of the costs given by Expression (1) is minimized. The graph cut unit 252 separates pixels corresponding to the cut subgraphs into groups 1 and 2, respectively.

  The separation unit 205 outputs the pixels separated into the groups 1 and 2 to the image generation unit 206.

  The image generation unit 206 performs a connected component analysis on each pixel of the two separated groups, and generates an image from any group of pixels. When any part of the group connected by the connected component analysis satisfies a condition for extracting a character, the image generation unit 206 generates a character image from the part that satisfies the condition.

  The condition for extracting characters can be determined by, for example, the aspect ratio and size of the connected component. This condition may be set in advance by examining the character-like size and aspect ratio by experiments or the like.

  As a result of the connected component analysis, if a connected component that seems to be a character is not obtained from any of the two groups, there is a possibility that the separation of the character and the background may have failed. It is only necessary to perform seed setting, graph generation, and graph cut on the density image having.

  The character image generated by the image generation unit 206 may be managed as character data by performing, for example, OCR (Optical Character Recognition) processing.

<Each process>
Next, specific examples of pattern extraction of the image processing apparatus 10 such as character extraction and image generation will be described.

(Channel decomposition processing)
The decomposition unit 202 performs a decomposition process on the image shown in FIG. For example, the decomposition unit 202 decomposes the image shown in FIG. 1 into a plurality of channels, and generates a density image corresponding to each channel. Note that the image shown in FIG. 1 is actually a color image, the background is light purple, and the characters are red.

  For example, the disassembling unit 202 disassembles into four channels of I channel, R channel, G channel, and B channel. Note that the number of channels to be decomposed is not limited to four.

  FIG. 9 is a diagram illustrating an example of the channel decomposition result. As shown in FIG. 9, the decomposition unit 202 decomposes the image into four channel density images. When the image is a gray image, the pixel value often represents a brightness value. The color pixel values of the three colors RGB can be converted into lightness values using, for example, a lightness value = 0.299 × R + 0.587 × G + 0.114 × B.

(Selection process)
The selection unit 203 calculates a density distribution (for example, a density histogram) for each density image, and calculates the maximum separation degree that maximizes the separation degree of the two groups for each channel. Next, the selection unit 203 calculates the maximum separation degree of the two groups for each channel based on the obtained density histogram, and determines the channel that maximizes the obtained maximum separation degree. This channel is also called the maximum resolution channel.

  The selection unit 203 may use, for example, a well-known Otsu system for the degree of separation of the two groups (for example, Noriyuki Otsu, “Automatic threshold selection method based on discrimination and least-square criterion”, IEICE paper) 80/4 Vol.J63-D No.4).

  For example, for the density histogram of the density image of interest, the selecting unit 203 sets the inter-group variance / in-group variance when the pixels are separated into two groups at each density value as the two groups, and the two groups are separated. The density at which the value is maximum is taken as the density threshold.

  The selection unit 203 sets the inter-group variance / in-group variance when the density histogram is separated into two groups by the density threshold as the maximum separation degree S of the target channel. The selection unit 203 obtains the maximum separation degree S for each channel, and determines the channel having the maximum S as the maximum separation degree channel.

  FIG. 10 is a diagram showing the maximum degree of separation in each channel. In the example shown in FIG. 10, the selection unit 203 obtains the maximum degree of separation for the density image of each channel shown in FIG. In the example shown in FIG. 10, the B channel is the maximum resolution channel. The selection unit 203 outputs the density image of the maximum resolution channel to the extraction unit 204.

(Extraction process)
The extraction unit 204 extracts the maximum density g ₁ and the minimum value g ₀ of the density image of the maximum resolution channel.

FIG. 11 is a diagram showing an example of the maximum value and the minimum value of the density image of the maximum resolution channel. As shown in FIG. 11, the extraction unit 204 sets the maximum value of the density image of the maximum resolution channel to g ₁ and sets the minimum value to g ₀ . The extraction unit 204 outputs the extracted maximum value and minimum value to the separation unit 205.

(Graph generation processing)
The graph generation unit 251 creates a graph with nodes corresponding to the groups 1 and 2 as end nodes as sink nodes, start nodes as source nodes, and image pixels as other nodes. The graph generation unit 251 establishes an N link by establishing a link between the target pixel and a node of a pixel in the vicinity thereof between other nodes.

  The graph generation unit 251 establishes links between other nodes and sink nodes, and other nodes and source nodes to form T links. The graph generation unit 251 assigns a smoothing cost to the N link by a known method. The smoothing cost is a cost that increases when the states (such as feature values) of neighboring pixels are similar and increases when they are different.

For example, when the density value of the pixel of interest is g, the graph generation unit 251 sets the distance between g and the maximum value g ₁ to d ₁ and sets the distance between g and the minimum value g ₀ to d ₀ . At this time, the graph generation unit 251 calculates the data cost of the T link between the sink node, the source node, and the target pixel according to the equations (4) and (5). Here, f represents a function.

  FIG. 12 is a diagram illustrating an example of a brightness histogram and a density histogram of a channel with the highest degree of separation. As shown in FIG. 12, even if the images are the same, when a brightness histogram is used, it may not be clear which group should be included for a pixel of a certain brightness. On the other hand, the density histogram of the channel with the highest degree of separation is a histogram in which the degree of separation of the two groups is maximized in the first place. Therefore, it can be easily determined whether a certain density of pixels is included in any group. Groups 1 and 2 correspond to either the foreground or the background.

  Here, the graph generation unit 251 may use the simplest expressions (6) and (7) for the expressions (4) and (5) as the T link.

Here, L represents the range that the density takes.

(Graph cut processing)
The graph cut unit 252 cuts the graph into two subgraphs with respect to the generated graph so that the total cost is minimized by, for example, the equation (1).

  As a simplest example, the graph cut unit 252 may obtain the energy obtained by the equation (8).

FIG. 13 is a diagram illustrating links and weights when attention is focused on the target pixel. The example shown in FIG. 13 shows an example of Formula (9) and Formula (10). In the example shown in FIG. 13, the density of the maximum resolution channel is used for A _i in equation (1). i is the number of a pixel in the image. The s node shown in FIG. 13 represents a source node, and the t node represents a sink node. The nodes n1 to n4 represent neighboring pixels (peripheral pixels).

  The T link weight represents the difference between the density of the target pixel and the density g selected as the true value, and the N link weight represents the difference between the target pixel and the density of the neighboring pixels. The larger the difference, the larger the energy.

  λ is set to an appropriate value through experiments or the like, and corresponds to the ratio of the T link weight and the N link weight.

  The graph cut unit 252 separates the pixels in the image into groups 1 and 2 depending on which subgraph corresponds to the pixel in the image. The graph cut unit 252 outputs the pixels included in the group 1 and the pixels included in the group 2 to the image generation unit 206.

(Image generation processing)
The image generation unit 206 determines whether one of the obtained two groups of pixels is a foreground by connected component analysis, and generates a character image from the foreground pixels. Furthermore, in order to use the generated character image for character recognition, the image generation unit 206 generates a binary image in which the character portion is black and the background is white.

  Further, the image generation unit 206 may generate a color image according to the application. FIG. 14 is a diagram illustrating an example of the generated binary image. The example shown in FIG. 14 is the result of performing a graph cut using the density image of the B channel and performing a connected component analysis. As shown in FIG. 14, character extraction is appropriately performed as compared with FIG. 2 and FIG.

  With the above configuration and function, in the embodiment, it is not necessary to manually set or learn the seed, and it is possible to automatically calculate the data cost of the T link.

<Operation>
Next, the operation of the image processing apparatus 10 in the embodiment will be described. FIG. 15 is a flowchart illustrating an example of image processing in the embodiment. In step S101 shown in FIG. 15, the decomposition unit 202 decomposes the channels of the input image and generates a density image for each channel.

  In step S102, the selection unit 203 (histogram calculation unit 231) calculates a density histogram for each channel.

  In step S103, the selection unit 203 (channel selection unit 232) calculates the maximum degree of separation for each channel.

  In step S <b> 104, the selection unit 203 (channel selection unit 232) selects the maximum separation channel having the largest maximum separation among the maximum separations for each channel.

  In step S105, the extraction unit 204 extracts the maximum value and the minimum value of the density of the maximum resolution channel.

  In step S106, the separation unit 205 (graph generation unit 251 and graph cut unit 252) generates a graph using the extracted maximum value and minimum value, and calculates the cost of the T link and the cost of the N link.

  In step S107, the separation unit 205 (graph cut unit 252) performs a graph cut so that the total cost is minimized, and separates each pixel of the density image into two groups.

  In step S <b> 108, the image generation unit 206 performs a connected component analysis on each group of pixels, and generates an image using a group of pixels that satisfy a condition for extracting characters. The generated image may be a binary image or a color image.

  As described above, according to the embodiment, character extraction can be appropriately performed. For example, according to the embodiment, it is possible to appropriately perform character extraction from an image obtained by imaging characters described in a medicine package, which has conventionally been difficult to extract characters.

  The above-described embodiment can be applied to an image including a pattern. However, the above-described embodiment functions particularly suitably for an image in which characters printed on a material that causes texture due to shadows and reflections are captured. . Moreover, in the said Example, it functions suitably for extracting the character printed on the package of a medicine, for example.

  In addition, by recording a program for realizing the image processing described in the above embodiment on a recording medium, the image processing in the above embodiment can be performed by a computer. For example, it is possible to record the program on a recording medium and cause the computer to read the recording medium on which the program is recorded, thereby realizing the above-described image processing.

  In addition, the image processing apparatus in the said Example is applicable if it is a processing apparatus which has processors, such as a computer (Personal Computer), a tablet terminal, a smart phone, memory, and an interface which acquires data. Further, the image processing apparatus 10 of the above-described embodiment may be included in the scanner apparatus so that the scanner apparatus can execute the image processing.

  The recording medium is a recording medium that records information optically, electrically, or magnetically, such as a CD-ROM, flexible disk, magneto-optical disk, etc., and information is electrically recorded such as ROM, flash memory, etc. Various types of recording media such as a semiconductor memory can be used. This recording medium does not include a carrier wave.

  Although the embodiments have been described in detail above, the invention is not limited to the specific embodiments, and various modifications and changes can be made within the scope described in the claims. It is also possible to combine all or a plurality of the components of the above-described embodiments.

In addition, the following additional remarks are disclosed regarding the above Example.
(Appendix 1)
A decomposition unit that decomposes the image into a plurality of channels and generates a density image of each channel;
A selection unit that separates the density distribution of each density image into two groups, and selects a density image that maximizes the degree of separation of the two groups;
An extraction unit for extracting the maximum value and the minimum value of the density of the selected density image;
A separation unit that separates the selected density image into two groups using a graph cut in which the extracted maximum and minimum values are set as seeds;
An image generation unit that performs a connected component analysis on each of the two separated pixels and generates an image from one of the pixels;
An image processing apparatus comprising:
(Appendix 2)
The selection unit includes:
A density histogram is generated from each density image, and for each density histogram, when the pixels are separated into two groups by each density value, the density between the groups is divided into the two groups with the maximum degree of separation divided by the intra-group variance. The image processing apparatus according to supplementary note 1, wherein the image processing apparatus separates and selects a density image having the largest degree of separation among the maximum degrees of separation of the density images.
(Appendix 3)
The image includes a character portion;
The image generation unit
The image processing apparatus according to claim 1 or 2, wherein when any part of the group connected by the connected component analysis satisfies a condition for extracting a character, a character image is generated from the part that satisfies the condition.
(Appendix 4)
Decompose the image into multiple channels and generate a density image for each channel,
The density distribution of each density image is separated into two groups, and the density image that maximizes the degree of separation of the two groups is selected,
Extract the maximum and minimum density values of the selected density image,
Using the graph cut set with the extracted maximum and minimum values as seeds, the selected density image is separated into two groups,
An image processing method in which a computer executes a process of performing a connected component analysis on each of two separated pixels and generating an image from one of the groups of pixels.
(Appendix 5)
Decompose the image into multiple channels and generate a density image for each channel,
The density distribution of each density image is separated into two groups, and the density image that maximizes the degree of separation of the two groups is selected,
Extract the maximum and minimum density values of the selected density image,
Using the graph cut set with the extracted maximum and minimum values as seeds, the selected density image is separated into two groups,
A program that causes a computer to perform a process of performing connected component analysis on each of the two separated pixels and generating an image from one of the groups of pixels.

DESCRIPTION OF SYMBOLS 10 Image processing apparatus 20 Imaging part 101 Control part 102 Main memory part 103 Auxiliary memory part 104 Communication part 105 Drive apparatus 201 Image input part 202 Decomposition part 203 Selection part 204 Extraction part 205 Separation part 206 Image generation part 231 Histogram calculation part 232 Channel Selection unit 251 Graph generation unit 252 Graph cut unit

Claims (5)

A decomposition unit that decomposes the image into a plurality of channels and generates a density image of each channel;
A selection unit that separates the density distribution of each density image into two groups, and selects a density image that maximizes the degree of separation of the two groups;
An extraction unit for extracting the maximum value and the minimum value of the density of the selected density image;
A separation unit that separates the selected density image into two groups using a graph cut in which the extracted maximum and minimum values are set as seeds;
An image generation unit that performs a connected component analysis on each of the two separated pixels and generates an image from one of the pixels;
An image processing apparatus comprising: The selection unit includes:
A density histogram is generated from each density image, and for each density histogram, when the pixels are separated into two groups by each density value, the density between the groups is divided into the two groups with the maximum degree of separation divided by the intra-group variance. The image processing apparatus according to claim 1, wherein the density image is separated and the density image having the largest degree of separation is selected from the maximum degrees of separation of the density images. The image includes a character portion;
The image generation unit
3. The image processing apparatus according to claim 1, wherein when any part of the group connected by the connected component analysis satisfies a condition for extracting a character, a character image is generated from the part that satisfies the condition. Decompose the image into multiple channels and generate a density image for each channel,
The density distribution of each density image is separated into two groups, and the density image that maximizes the degree of separation of the two groups is selected,
Extract the maximum and minimum density values of the selected density image,
Using the graph cut set with the extracted maximum and minimum values as seeds, the selected density image is separated into two groups,
An image processing method in which a computer executes a process of performing a connected component analysis on each of two separated pixels and generating an image from one of the groups of pixels. Decompose the image into multiple channels and generate a density image for each channel,
The density distribution of each density image is separated into two groups, and the density image that maximizes the degree of separation of the two groups is selected,
Extract the maximum and minimum density values of the selected density image,
Using the graph cut set with the extracted maximum and minimum values as seeds, the selected density image is separated into two groups,
A program that causes a computer to perform a process of performing connected component analysis on each of the two separated pixels and generating an image from one of the groups of pixels.