JP4322310B2 - Color pattern extraction apparatus and color pattern extraction program - Google Patents

Description

  The present invention relates to a color pattern extraction apparatus that extracts a pattern of a specific color from a color image, and a color pattern extraction program that causes a computer to function as such an apparatus, and more particularly to a pattern such as a stamp post-printed on color paper. It is related with the technique suitable for extracting.

  A technique for extracting a pattern of a specific color from a color image has been conventionally used, for example, for extracting a pattern such as a stamp from a color form on which a pattern such as a stamp or a seal is post-printed. For example, a method is known in which the color of a post-printed pattern is registered, and a color pattern that approximates the registered color is extracted from an image obtained by optically reading a form or the like. However, even if the color of the pattern formed by ink such as a stamp is the same, if the background color, which is the color of the paper, is different, it may be mixed with the background color and become a different color. is there. If the color of the pattern in the image read in this way changes due to color mixing, the approximation with the color of the registered pattern is lost and the pattern may not be extracted.

  On the other hand, Patent Document 1 discloses a method of comparing a pattern color in an image with a registered color on the basis of the background color in the image in order to remove the influence of the color mixture with the background color. In this method, the angle θc formed by the color vector of the target pattern in the image and the color vector of the background color, and the angle θt formed by the color vector of the registered color and the color vector of the background color in the image are calculated. When the angle difference is smaller than a predetermined threshold, it is determined that the color of the target pattern matches the registered color. By calculating the angle θt for a plurality of registered colors, the color of the target pattern can be assigned to the registered color with the smallest angle difference, and the pattern corresponding to the designated color can be extracted.

Japanese Patent Laid-Open No. 10-104075

  However, in the method of Patent Document 1, if the degree of color mixture between the pattern in the image and the background is large, and the color shift between the pattern color in the image and the registered color increases, the angle difference increases, and the color determination There is a high possibility that pattern extraction will fail.

  Further, in the method of Patent Document 1, when a pattern of colors having the same color system and different RGB values is mixed in an image, there is a possibility that a pattern of a color different from that designated by the registered color may be extracted. For example, if there are stamps and characters of the same color but different RGB values on a chromatic area that is different from the basic background color at the time of registration, and only the post-printed stamp pattern is to be extracted, it is represented by ink. The stamp color is different from the stamp color on the basic background color registered by the color mixture with the background color. At this time, if the color of the printed character is close to the stamp color on the registered basic background color, the character pattern is erroneously extracted.

  The present invention has been made in view of the above problems, and the problem is to accurately extract a pattern based on a registered color even when the color of the target pattern changes greatly due to color mixing with the background. Another object of the present invention is to provide a color pattern extraction apparatus capable of performing the above and a color pattern extraction program that causes a computer to function as such an apparatus.

  In order to solve the above problems, a color pattern extraction apparatus according to the present invention is an apparatus that extracts a specific color pattern from a color image, and uses a color on the basic background color of the specific color pattern to be extracted. A set of a basic pattern color and a basic background color, and at least two sets of a reference pattern color and a reference background color, which are colors on a reference background color different from the basic background color of a specific color pattern Based on information, a correction parameter for converting a pattern color obtained when a pattern of a specific color is displayed on an arbitrary background color into a color estimated to be obtained when displayed on a basic background color The target pattern is obtained using the learning means for obtaining the target pattern color and the background color by separating the input image into the target pattern and the background, and using these values and the correction parameter obtained by the learning means. Is converted to a color that is estimated to be obtained upon display on a basic background color, characterized in that it comprises an extraction means for extracting a matching pattern from the transformed object pattern to the basic pattern color.

  The learning means includes a learning pattern registration unit that registers a basic image in which a pattern of a specific color is represented on a basic background color and a reference image in which a pattern of a specific color is represented on a reference background color, a basic image, and a reference For each of the images, a separation unit that separates the pattern and the background, a color calculation unit that obtains a representative color of the separated pattern and the background, a correction parameter setting unit that obtains a correction parameter based on the obtained representative color, It is desirable to provide.

  In addition, the extraction unit includes a separation unit that separates the input image into a target pattern and a background, a color calculation unit that obtains a representative color of the separated target pattern and the background, and the obtained representative color and correction parameter. It is desirable to include a color mixture correction unit that converts the representative color of the target pattern based on the above and a determination unit that determines whether or not the color of the target pattern subjected to color conversion matches the basic pattern color.

  As a color overlap model in the learning means and extraction means, a particle density model is applied, and at least for the pattern color, a parameter is introduced into the density coefficient of the particle density model to convert the difference in color for each part of the pattern. It is desirable to be reflected in

  On the other hand, the color pattern extraction program according to the present invention allows a computer to set a combination of a basic pattern color and a basic background color that is a color on a basic background color of a specific color pattern to be extracted, and a specific color pattern. A pattern of a specific color is displayed on an arbitrary background color based on color information of at least two sets of a reference pattern color and a reference background color, which is a color on a reference background color different from the basic background color of A learning function to obtain correction parameters for converting the pattern color obtained in this case into a color estimated to be obtained when displayed on the basic background color, and separating the input image into the target pattern and the background The target pattern color and the background color are obtained, and the estimated color obtained when the target pattern color is displayed on the basic background color using these values and the correction parameter obtained by the learning means. Converted to an extracting function of extracting a pattern matching the basic pattern color from the converted target pattern, characterized in that for the execution.

  According to the color pattern extraction apparatus of the present invention configured as described above, even when the color of the pattern post-printed on the background color changes greatly due to color mixing with the background color, the pattern color is used as the basic background. It is possible to correct the color when it is displayed on the color, and it is possible to accurately execute the matching determination with the designated color and extract the color pattern.

  Embodiments of a color pattern extraction apparatus and a color pattern extraction program according to the present invention will be described below with reference to the drawings. First, an overview of the entire system including the color pattern extraction apparatus of the present embodiment will be described based on FIG. As shown in FIG. 1, the system 1 according to the embodiment includes a color pattern extraction device 10, a display 30, a printer 31, and a scanner 32 connected to the color pattern extraction device 10.

  The color pattern extraction apparatus 10 is configured by a general-purpose computer such as a personal computer, and includes a CPU 11, a hard disk (HD) 20, a memory (RAM) 12, and an interface 13 connected to the CPU 11. An operating system (not shown) is installed in the HD 20 and a color pattern extraction program 21 is installed. The display 30, the printer 31, and the scanner 32 are connected to the CPU 11 via the interface 13. The CPU 11 develops and executes the color pattern extraction program 21 on the RAM 12 as necessary. The target color image is read by the scanner 32 or provided as an image file. Further, the extraction process and results are displayed on the display 30 as necessary, and printed out by the printer 31.

  The color pattern extraction program 21 includes a learning module 22 that registers a pattern and a background color as extraction pre-processing and obtains a correction parameter, and an extraction module 23 that extracts a pattern from an actual image.

  The learning module (learning means) 22 is a combination of a basic pattern color and a basic background color, which is a color on a basic background color of a specific color pattern to be extracted, and a basic background color of the specific color pattern. A pattern color obtained when a specific color pattern is displayed on an arbitrary background color based on at least two sets of color information of a reference pattern color and a reference background color, which are colors on different reference background colors Has a function for obtaining a correction parameter for converting to a color estimated to be obtained when the image is displayed on the basic background color. In the learning module 22, a learning pattern registration unit 22a that registers a basic image in which a specific color pattern is represented on the basic background color and a reference image in which the specific color pattern is represented on the reference background color; For each of the basic image and the reference image, a separation unit 22b that separates the pattern and the background, a color calculation unit 22c that calculates a representative color of the separated pattern and the background, and correction based on the obtained representative color And a correction parameter setting unit 22d for obtaining parameters. The representative color is one in which one color is assigned to each pattern, for example, by averaging the colors of the pixels constituting the pattern. Each color is obtained by, for example, taking an average value after extracting a pattern.

  On the other hand, the extraction module (extraction means) 23 obtains the target pattern color and the background color by separating the input image into the target pattern and the background, and uses these values and the correction parameter obtained by the learning module 22. The target pattern color is converted into a color estimated to be obtained when displayed on the basic background color, and a pattern matching the basic pattern color is extracted from the converted target pattern. The extraction module 23 includes a separation unit 23a that separates the input image into a target pattern and a background, a connected component calculation unit 23b that separates the target pattern for each connected component, and representatives of the background color and each connected component. The color calculation unit 23c for obtaining a color, the color mixing correction unit 23d for converting the representative color of the target pattern based on the obtained representative color and the correction parameter, and the color of the color-converted target pattern matches the basic pattern color And a determination unit 23e for determining whether or not. The determination unit 23e determines a match between the corrected target pattern and the designated basic pattern color using a known method. For example, the case where the angle between the color vector of the target pattern after correction and the color vector of the basic pattern color is equal to or smaller than a threshold value is determined to be coincident, and the case where the angle exceeds the threshold value is determined to be inconsistent.

  Next, the extraction principle of the color pattern extraction apparatus of the present invention will be described based on specific examples. In the following description, an image obtained by reading a paper on which a handwritten pattern with ink is post-printed by the scanner 32 is a processing target, a paper color is a background color, and a post-printed handwritten pattern is an extraction target. The basic background color is white.

As shown in FIG. 2A, the representative color of the target pattern (handwritten A) post-printed with a specific color ink on the basic background color S ₀ is C _A. C _A corresponds to the basic pattern color. In addition, as shown in FIG. 2 (B), the representative color of the target pattern (handwritten ABCD) that is post-printed with the same specific color ink as above on a color form with a reference background color S ₁ different from the basic background color. Is C _A '. C _A 'corresponds to the reference pattern color. In FIG. 2B, the pattern “item name” of the representative color D that is not the extraction target is printed in the same image.

The learning module 22 obtains a correction parameter based on the two sets of color information (S0, CA) and (S1, CA ′). In this example, a known particle density model (Terai et al., “Base Color Recognition by Tetragonal Regression for Overlapped Watercolors”, Proc. Of IWFHR-9, 2004) is used as a color mixture model for defining color overlap. When this model is used, the colors C _A and C _A ^′ of the target pattern post-printed on the background colors S ₀ and S ₁ are expressed as equations (1) and (2), respectively. Since S ₀ is a basic background color (white), C _A ^′ is a mixed color of C _A and S ₁ as shown in Expression (3).

In the above equation, T _A is the ink color (specific color) of the target pattern, ρ _A is a coefficient corresponding to the ink density of the target pattern, and 0.0 when the color is uniform with respect to the target color surface. ≦ ρ _A ≦ 1.0 can be considered as a constant. As described above, if two sets of background color and pattern color are known, ρ _A can be obtained. If ρ _A is determined, it is estimated that it can be obtained when the color of the pattern that has undergone post-printing with a specific color ink on any background color is corrected and displayed on the basic background color. Can be converted (mixed color correction). ρ _A is generally determined by the color distribution (not the representative color of the pattern) of each pixel constituting the pattern.

For example, when color mixture correction is performed using the image of FIG. 2B as the target image, the color C _A ' ⁺ , which is estimated to be obtained when the pattern colors C _A ', D are displayed on the basic background color S ₀ , The image is converted to D ⁺ and an image as shown in FIG. 2C is obtained. 3A is a vector representing the background color and pattern color of the image of FIG. 2B, and FIG. 3B is the background color and pattern of the image of FIG. 2C after color mixture correction. It is a vector which shows the representative color of a color.

Here, FIG. 3 (A), an angle theta ₁ formed by the color vectors of the color D of the non-target pattern "item name" for color vectors of the basic pattern color C _A is the color vector of the basic pattern color C _A On the other hand, a case is assumed where the color vector of the target pattern (handwritten ABCD) is smaller than the angle θ ₂ . In this case, if it is attempted to extract a pattern without performing color mixture correction in the state shown in FIG. 3A as in the prior art, if the threshold value for determining the same color is θ ₁ , as shown in FIG. 4A. only non-target pattern is extracted, the pattern to be extracted originally coming off, when the threshold theta _2, not only the target pattern as shown in FIG. 4 (B), of the non-target patterns are extracted.

On the other hand, when executing the color mixing correction, as shown in FIG. 3 (B), the color of the target pattern approaches 'to C _A' C _A to a specific color are converted to ⁺ basic pattern color C _A, of the non-target pattern The color is converted from D to D ⁺ and away from the basic pattern color C _A. As a result, the angle theta ₁ ⁺ formed by the color D ⁺ color vector after the correction of the non-target patterns for color vectors of the basic pattern color C _A is the target pattern for the color vector of the basic pattern color C _A color It is larger than the angle θ ₂ ⁺ formed by the vector. For this reason, when the threshold value for determining that the colors are the same is θ ₂ ⁺ , only the target pattern is extracted as shown in FIG.

When the target pattern is a line drawing of characters or the like, the color distribution of each pixel constituting the line pattern has an extremely high variance in one axial direction, a distribution form on the line, and ρ _A is The assumption that it is a constant is difficult to hold. However, it can be assumed that the background portion is substantially uniform, as is the case with the surface figure. An example of the color distribution for each pixel when the pattern is a line drawing will be described with reference to FIG. FIG. 5 shows a color distribution E of each pixel constituting a pattern post-printed on the basic background color, and each pixel constituting a pattern post-printed with the same ink on a predetermined background color different from the basic color. The color distribution F, the color distribution G of each background pixel of the basic background color, and the color distribution H of each background pixel of the predetermined background color are shown. The color mixture correction means conversion from F to G.

Therefore, in the embodiment, a parameter t representing the variation of ρ _A is introduced and replaced as shown in Equation (4). The range of t is as shown in Equation (5).

Further, ρ _A (0) = 0 and ρ _A (1) = 1 are set. Therefore, the equations (1) to (3) can be rewritten as the following equations (6) to (8).

  From equations (6) and (7), equations (10) and (11) are established at t = 1 and t = 0, respectively.

Therefore, T _A can be obtained by obtaining two approximate straight lines of C _A (t) -S ₀ and C _A ′ (t) -S ₁ from the respective distributions and obtaining the intersection coordinates of the two straight lines. C _A ′ (t) is observed for a certain pixel, and t = t _p at this time. The following formula (12) is derived from the formula (7). On the other hand, the background colors are approximated as shown in Expression (13) by using the distribution average value.

When equation (14) is derived from the above equation and equation (8), and this is used to display this on the basic background color based on the pattern color C _A '(t) of the input image The color C _A (t) can be estimated.

From the above derivation, it can be seen that by preparing two combinations of the background color and the pattern color on the background color, the pattern color on an arbitrary background color can be restored. The learning module 22 learns the above T _A from the two sets of colors.

Here, the difference between the particle density model of Terai et al. Published in the aforementioned paper and the extended particle density model of the present application in which a parameter is introduced into ρ _A will be conceptually described with reference to FIGS. 6 and 7. The color distribution of the background pixels is assumed to be almost uniform in all models, and is indicated by S ₀ and S ₁ . On the other hand, the color distribution of each pixel of the pattern post-printed on the background color is assumed to be almost uniform in Terai et al.'S particle density model and is indicated by C _A and C _A ′. In the extended particle density model, by introducing a parameter to ρ _A , the distribution is assumed to be a straight line in consideration of the actual distribution.

If you try to convert the color distribution of a line drawing pattern printed on a background color different from the basic background color in Terai et al. To a color on the basic background color, the correct distribution is shown in Fig. 8. Not converted. Therefore, there is a possibility that the line drawing pattern cannot be extracted correctly. On the other hand, according to the model of the present application, by calculating an approximate straight line of C _A (t) -S ₀ and C _A ′ (t) -S ₁ , the background color different from the basic background color is obtained. It is possible to correctly convert the color distribution of the line drawing pattern post-printed to the color distribution estimated to be obtained when displayed on the basic background color. Therefore, when extracting a line drawing pattern such as characters, the extraction accuracy can be improved by using the extended particle density model of the present application.

  Next, the flow of processing of the color pattern extraction program 21 of the embodiment will be described based on a flowchart. When the process of the learning module 22 starts, as shown in FIG. 9, a learning image pattern is registered by the action of the learning pattern registration unit 22a (S01). Here, an image of the target pattern post-printed on the basic background color to be extracted and an image of the target pattern post-printed with the same ink on a reference background color different from the basic background color are registered. FIG. 10A shows an example of an image to be registered.

  Next, each of the registered two sets of images is binarized by an existing method and separated into a pattern and a background by the action of the separation unit 22b (S02). FIG. 10B shows a binary image of a pattern separated from the background. The color calculation unit 22c then displays the representative color of the basic background color and the color of each pixel, the representative color of the reference background color and the color of each pixel, the representative color of the target pattern post-printed on each background color, and each pixel. The color of is calculated (S03). FIG. 10C shows an example in which a background and a pattern are displayed in respective representative colors for each image.

  The correction parameter setting unit 22d calculates the parameters in the color mixture correction formula of the target pattern according to the above extended particle density model (S04). The representative color of the target pattern on the basic background color registered by the learning pattern registration unit 22a is registered as a color that can be selected when the user specifies at the time of pattern extraction, and the basic background color and the correction parameter are used for correction calculation. Is registered in the parameter file as a coefficient.

  When the process of the extraction module 23 starts, the user selects and specifies a pattern color (extracted ink color) to be extracted from the colors registered in the process of the learning module 22 (S11). Subsequently, an extraction target image in which a pattern is post-printed with the same ink on an arbitrary background color is input (S12). An example of the input image is shown in FIG. The separation unit 23a separates the image input by binarization into a pattern and a background (S13). A binary image of the separated pattern is shown in FIG.

  The connected component calculation unit 23b performs labeling on the binary image and extracts the connected component from the pattern (S14). FIG. 12C shows a calculation result of a connected component, and a portion surrounded by a dotted circumscribed rectangle is one connected component. Subsequently, the color calculation unit 23c calculates a representative color of the background color and a representative color of each connected component of the pattern (S15). FIG. 12D shows a representative color for each connected component.

  The color mixture correction unit 23d uses the correction formula obtained by the learning module 22 to perform color mixture correction for each connected component using the background color and the representative color of the connected component, and estimates the color when displayed on the basic background color. (S16). FIG. 12E shows the corrected pattern. The determination unit 23e collates the representative color after the color mixture correction of each connected component with the designated color (representative color of the pattern for which extraction is designated) (S17), and whether or not the corrected pattern color matches the designated color. As shown in FIG. 13, it is determined for each connected component by threshold processing (S18). If it is determined that they match, the target connected component is extracted as a target pattern (S19), and if it is determined that they do not match, the target connected component is deleted (S20). Note that if it is desired to make a determination in units smaller than the connected components, for example, in units of pixels, the color mixture correction and determination of matching with the specified color may be performed in units of pixels without performing the connected component calculation in S14.

  The extraction result is displayed on the display 30 or printed out by the printer 31 according to the user's request. At that time, a color image (shown in gray scale in the figure) may be output as shown in FIG. 14A, or a binary image may be output as shown in FIG. 14B.

1 is a block diagram showing a system including a color pattern extraction apparatus according to an embodiment of the present invention. (A) is an image that has been printed with a specific color ink on the basic background color, (B) is an image that has been printed with a specific color ink on the reference background color, and (C) is an image of (B). It is explanatory drawing which shows the image obtained by carrying out color mixing correction | amendment with the apparatus of embodiment. (A) is a color vector of the image of FIG. 2 (B), (B) is explanatory drawing which shows the color vector of the image of FIG.2 (C). (A) and (B) are explanatory diagrams showing extraction results by a conventional extraction method, and (C) is an explanatory diagram showing extraction results by the apparatus of the embodiment. It is a graph which shows the example of the color distribution for every pixel in case a pattern is a line drawing. It is explanatory drawing which shows notionally the color mixture correction | amendment by a particle density model of Terai. It is explanatory drawing which shows notionally the color mixing correction | amendment by the extended particle density model of embodiment. It is explanatory drawing which shows notionally the result of carrying out color mixture correction | amendment of the color distribution of a line drawing pattern by Terai et al. It is a flowchart which shows the processing content of the learning module of the color pattern extraction apparatus of embodiment. 9 is an example of the image processed in FIG. 9, (A) is a registered image, (B) is a binary image separated from the background, and (C) is an image displaying the background and the pattern in respective representative colors. It is explanatory drawing shown. It is a flowchart which shows the processing content of the extraction module of the color pattern extraction apparatus of embodiment. FIG. 11 is an example of an image processed in FIG. 11, (A) is an input image, (B) is a binary image of a separated pattern, (C) is a calculation result of connected components, and (D) is for each connected component. A representative color (E) is an explanatory diagram showing a pattern after color mixture correction. It is explanatory drawing which shows the mode of the threshold value process for every connected component in the process of FIG. (A) is an image representing a color image of the extraction result in gray scale, and (B) is an explanatory view showing a binary image of the extraction result.

Explanation of symbols

10 Color pattern extraction device 11 CPU
13 RAM
14 interface 20 HD
21 Color pattern extraction program 22 Learning unit 23 Extraction unit

Claims (5)

A color pattern extraction apparatus for extracting a specific color pattern from a color image,
On the reference background color that is different from the basic background color of the pattern of the specific color and the basic background color of the basic pattern color that is a color on the basic background color of the pattern of the specific color to be extracted A pattern color obtained when a pattern of the specific color is displayed on an arbitrary background color based on at least two sets of color information of a reference pattern color that is a color of the reference color and the reference background color. Learning means for obtaining a correction parameter for conversion to a color estimated to be obtained when displayed on a background color;
The input image is separated into the target pattern and the background to obtain the target pattern color and the background color, and using these values and the correction parameter obtained by the learning means, the target pattern color is determined as the basic background color. An extraction means for converting to a color estimated to be obtained when displayed above, and extracting a pattern that matches the basic pattern color from the converted target pattern;
A color pattern extraction apparatus comprising: The learning means includes
A learning pattern registration unit for registering a basic image in which the pattern of the specific color is represented on the basic background color and a reference image in which the pattern of the specific color is represented on the reference background color;
For each of the basic image and the reference image, a separation unit that separates a pattern and a background;
A color calculation unit for obtaining a representative color of the separated pattern and the background;
A correction parameter setting unit for obtaining the correction parameter based on the obtained representative color;
The color pattern extraction apparatus according to claim 1, further comprising: The extraction means includes
A separation unit that separates an input image into a target pattern and a background;
A color calculator for obtaining representative colors of the separated target pattern and the background;
A color mixing correction unit that converts a representative color of the target pattern based on the obtained representative color and the correction parameter;
A determination unit that determines whether or not the color of the target pattern subjected to color conversion matches the basic pattern color;
The color pattern extraction apparatus according to claim 1, further comprising:   A particle density model is applied as a color overlap model in the learning unit and the extraction unit, and at least for the color of the pattern, a parameter is introduced into the density coefficient of the particle density model, and the color difference for each part of the pattern The color pattern extracting apparatus according to claim 1, wherein the color pattern is reflected in color conversion. A color pattern extraction program that causes a computer to operate yesterday as a color pattern extraction device that extracts a pattern of a specific color from a color image,
On the computer,
On the reference background color that is different from the basic background color of the pattern of the specific color and the basic background color of the basic pattern color that is a color on the basic background color of the pattern of the specific color to be extracted A pattern color obtained when a pattern of the specific color is displayed on an arbitrary background color based on at least two sets of color information of a reference pattern color that is a color of the reference color and the reference background color. A learning function for obtaining a correction parameter for conversion to a color estimated to be obtained when displayed on a background color;
The input image is separated into the target pattern and the background to obtain the target pattern color and the background color, and using these values and the correction parameter obtained by the learning means, the target pattern color is determined as the basic background color. An extraction function that converts to a color estimated to be obtained when displayed above and extracts a pattern that matches the basic pattern color from the converted target pattern;
A color pattern extraction program characterized in that