JPS6252675A - Method for extracting specific color area - Google Patents

Abstract

PURPOSE:To improve accuracy in processing and analysis by converting level information at every channel into vector information representing color attributes (color brightness, hue and saturation degree). CONSTITUTION:The output of an primitive signal generator 1 is applied to an RGB, an I(brightness) converter 2, an H(hue) converter 3 for RGB to H(hue) and a converter 4 from RGB to P(saturation degree) through digital lines 1R, 1G and 1B. Outputs from respective converters 2-4, that is, one-dimensional data equivalent to brightness, hue and saturation degree and the corresponding reference levels of said brightness, hue and saturation degree are impressed to window comparators 5i, 5h and 5p, and only areas having color attributes in a desired area are made 'true' among original pictures in terms of their outputs. The outputs of the comparators are all impressed on an AND circuit 6. Accordingly all the outputs consider color attributions.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for extracting a specific region of an original image in image processing or (and) image analysis, or for identifying the color of a specific region. Related to area extraction method.

Traditionally, extraction and identification of specific areas in images generally involve
The main method used was to use a monochrome image as the original image and perform processing based on density information using a simple comparator.

However, in various measurements that apply image analysis, such as the analysis of chromosomes and cells, and the measurement of impurities in metals (non-metallic inclusions), color analysis is required due to the large amount of information and to improve the accuracy of analysis. The demand for using images is increasing year by year. As a processing technique using such a color image, for example, Japanese Patent Application Laid-open No. 57-79, which the present applicant previously applied,
There are methods such as those described in No. 944 and Japanese Unexamined Patent Publication No. 58-22929.

In conventional image processing for color images, for example, in the method described in Japanese Patent Application Laid-open No. 79944/1987, the absolute levels of each channel of red, green, and blue are processed as they are in the original state. Since it is used as data, its practical discrimination ability is significantly inferior when the original image deviates from the ideal state due to shading or the like. Furthermore, if you intentionally expand the discrimination region, for example when you want to extract a certain hue with a slight change in brightness as the same area, there is a drawback that the hue to be discriminated will also be greatly enlarged as a side effect. Was. Furthermore, in the method described in JP-A No. 58-22929, the concept of color discrimination is not perfect because it targets only hue among color attributes, and there is no concept of a hue circle, that is, a vector, so there are two sets of limitations. Since parameters are used, it is difficult to establish correspondence with the concept of color, and furthermore, it is essentially impossible to make settings when there is no desired color tone in the observation screen.In other words, extrapolation settings are not possible regardless of the actual image. It has many drawbacks such as being unable to do so.

SUMMARY OF THE INVENTION In view of the current situation, the present invention aims to provide a method for extracting a specific color region that is extremely practical for color processing of color images and that improves the accuracy of processing and analysis. do.

3yIυU In order to extract or identify a specific color from a multiband image recorded in at least two or more channels, the present invention uses level information for each channel to extract color attributes (brightness, hue, saturation). ) into representative vector information and perform discrimination based on this information.

More specifically, the present invention provides the following method. That is, (1) As a means of converting into brightness axis components, use the sum of the absolute values of each band multiplied by appropriate coefficients,
Furthermore, as a means of conversion into hue axis components, the angle formed by the composite vector of a group of vectors representative of the level of each channel, which are placed at appropriate angular intervals on the orthogonal coordinates, with the reference axis is used, and when calculating this angle, each channel is The cosine component and sine of the composite vector are calculated using a coefficient unit that calculates the cosine component of the angle that each representative vector forms with the reference axis of the orthogonal coordinates, a coefficient unit that calculates the sine component, and an adder that calculates the sum of each component. A method that uses the arctangent with the components as input, and the output of the calculator.

(2) As a means of changing the lightness axis component, use the sum of the absolute values of each band multiplied by an appropriate coefficient,
Furthermore, as a means of changing the saturation axis component, the absolute value of the composite vector is changed to 1■,! The absolute value of the composite belt whose elements are only the two largest vectors representing the level of each channel recording the original image is IVP +
Therefore, l Vll l / l VP
A method that uses the output of an arithmetic unit that performs the operation I.

(3) As a means of changing the lightness axis component, use the sum of the absolute values of each band multiplied by an appropriate coefficient,
Furthermore, as a means of changing the saturation axis component, let the absolute value of the composite belt be |VH|, and let the sum or average of the levels for each channel recording the original image be L; IV)
A method using the output of an arithmetic unit that performs the calculation 11/L.
- The present invention will be described in detail below, but in order to better understand the present invention, in the following explanation, multi-band images will be described as a representative example of three bands consisting of red (R), green (G), and blue (B). We will focus on primary color images, and unless there is a risk of confusion, each will be expressed using the symbols RlG and B.

In addition, color attributes such as lightness (I), hue (H), and saturation (P) are expressed by symbols I and HIP unless there is a risk of confusion.

First, before detailed explanation of the present invention, the following matters will be described as a premise.

1) Extraction of a specific region of an image (for example, a monochrome image) that has only one channel as an element is
Since the setting parameters are one-dimensional, it is easier to do this using well-known means than in the past, but it is naturally also possible using the method of the present invention. .

2) Three-dimensional parameters are required to extract a specific color region from a three-channel image consisting of two or more elements, for example R, G, and B. This means that even when focusing on only one of the three-dimensional vectors representing color attributes, ie, I, H, and P, all three-dimensional parameters must be considered.

3) The three-dimensional original data of R, G, and H is converted into ■, H,
If the color attribute is converted to P, each one-dimensional parameter can be used to easily identify a specific area.

4) The logical product of the three extracted specific areas related to each color attribute of I, H, and P is completely equal to the specific color area obtained by specifying all of I, H, and P from the beginning. This is because when considering a color solid, specifying a specific color means determining one point within it, but specifying ■ means determining one point in the vertical direction of the color solid.
The color you are considering will exist somewhere on the plane when the color solid is sliced horizontally at this location. Specifying H here means specifying the angle from the reference axis when this plane is considered to be orthogonal coordinates with the achromatic axis of the color solid as the origin. Therefore, the color we are considering exists somewhere on the line segment representing this angle. Furthermore, specifying P is equivalent to specifying the diameter of a concentric circle centered on the achromatic axis of the color solid, so one point on this line segment is determined.

5) If it is sufficient to pay attention to only one of ■, H, and P, it is sufficient to consider only the parameters related to it.

6) The linearity in the conversion process from R,, G, B to ■, H,, P is not important from the perspective of the field of application of the present invention, and furthermore, the absolute value of the conversion is not important.

What is required is monotonic
y) and continuity, that is, I, H4F, the combination of RlG, B, which is the original data, is fixed to only one.

To summarize the above points, in specific color region extraction, R,
The original data consisting of three dimensions G and B is I, ■1
, P, the subsequent handling becomes extremely easy.

The present invention has been made based on the ideas 1) to 6) above, and its basic configuration is straightforward and clear as shown in FIG.

Although the present invention can be applied to both analog and digital signal systems in exactly the same way, the digital system will be used for explanation here.

Explanation with reference to the drawings Now, the primitive signal generator 1 in Fig. 1 provides a color image expressed by RlG,B, and is generally an A/D converter of the output of an RGB color camera. ,
Alternatively, it is output from the image memory.

The output of the original signal generator 1 is applied to an RGB-1 converter 2, an RGB-H converter 3, and an RGB-P converter 4 through digital lines IRlIG and IB. Code 5i, 5h
The output from each converter 2.3.4, that is, the one-dimensional data corresponding to I, H, and P, is added to the window comparators indicated by , 5p, respectively. . Further, all the outputs of each comparator are applied to the OR product circuit 6. Therefore, the output cuff is 4 on page 9 of this specification.
), all color attributes are considered.

Here, the method of applying each reference level related to I, H, and P will be explained.
929, a method of sampling representative points in the image, a method of extrapolating a predicted level in advance, or a method of observing the extraction area on a display while changing it continuously. Alternatively, a method of determining by trial and error may be used. However, the method of applying the reference level itself is not directly related to the gist of the present invention.

Next, each transducer of FIG. 1, which forms the gist of the present invention, namely R
GB-I converter 2, RGB-+H converter 3, RGB-P
The converter will be specifically explained.

First, the RGB-I converter 2 will be explained. Since this can be easily determined by multiplying each channel by an appropriate coefficient and then calculating the sum, it is not particularly important in the present invention. Here, the appropriate coefficient is determined by taking into consideration the difference in human visual intensity and the reproduction level of each channel of the device, or the weighting coefficient for each channel depending on the purpose of analysis. It is.

Next, the most distinctive feature of the present invention is RGB.
The constituent means of the -H converter 3 will be explained below.

As is well known, if hue is expressed on a plane, 1
form two rings. Therefore, if this ring is placed on orthogonal coordinates, all hues can be expressed as angles from a reference axis, such as the X axis, on the orthogonal coordinates. This situation is shown in FIG. As you can see from this example of the color wheel, R
, G, and B can be expressed using one-dimensional data called angle. Therefore, the minimum value of data that can be held by each channel of R, G, and B can be expressed by using one-dimensional data called angle. If we consider zero and maximum values as normalized values, the hues that can be expressed using R, G, and B channels by a device that provides an image from which a specific color region is to be extracted are shown in Figure 3. 6 with red (R), yellow (Y), green (G), cyan (C), blue (B), and magenta (M) as the vertices.
It is considered to be inside a polygon (this is an example of a hexagon drawn by a composite vector when the original data RSG, B is considered as a vector), and wherever it is, it can be expressed as an angle from the reference axis. Can be done.

In other words, when each level of the R, G, and B channels is considered as a vector and placed on a plane at arbitrary angles (for example, at equal intervals), the hue is continuous as the angle from the arbitrary reference position indicated by the composite vector. It can be expressed as a monotonically increasing one-dimensional quantity.

In the present invention, for the sake of simplicity, the reference axis will be referred to as the horizontal axis (
AX) in FIGS. 2 and 3, the R vector is made to coincide with AX, and the G vector and B vector are arranged at equal intervals.

When this is done, the RGB shown in FIG. 1 becomes the present invention.
-H converter 3 can be constructed as shown in FIG.

In Figure 4, GX of the coefficient multiplier group is the horizontal component of the G vector, GY is the vertical component, BX is the horizontal component of the B vector, B
Y is the same vertical component, and RX is the horizontal component of the R vector.

However, the R vector has no vertical component. Each of these components is combined by an adder 10.11 to create a horizontal axis component CX and a vertical component CY of the combined vector. Further C
X and CY are converted into angles of a composite vector via an arctangent calculator 12. Table 1 shows an example of a correspondence table when hue is converted into angle in this way. This table shows an example of the numerical representation of hue having the configuration shown in FIG.

In Table 1, angles are expressed by replacing them with simple numerical values, with 360' being expressed as a numerical value of 256 (=2''), and other angles expressed by proportional distribution.

Next, the RGB→P converter 4 in FIG. 1 will be specifically explained, but the present invention is based on the following idea in RGB-P conversion. That is, 1) A color with 100% saturation is expressed using only two pieces of original data at most. For example, R only, R and G mixed in an appropriate ratio, C (cyan) mixed equally with G and B, etc. all have saturation (or purity).
It is 100χ.

2) A color in which all original data are mixed equally has a saturation level of 0%.

Now, FIG. 5 shows the configuration of an embodiment of the RGB-P converter according to the present invention.

The maximum value selector 21 in FIG.
The largest one is selected from B (if there are two or more largest ones, any of them may be used), and its output M1 is input to the vector sum calculator 22.

On the other hand, the next largest value selector 23 selects the remaining original data R, G excluding the one corresponding to Ml among R, G, and B.

The largest one of B (if there are two or more largest ones, any of them may be used) is selected and inputted to the vector sum calculator 22 as M2. The vector sum calculator 22 considers these two input data as vectors, and calculates Ml
, M2 is RlG or B, respectively.
The horizontal component vh and vertical component Vv of the composite vector are calculated by the same procedure as in the GB-H converter 3 (FIG. 1). These vh and Vv are given to the vector length calculator 24, which calculates the absolute value IVMI. Needless to say, at this time, computation is equivalently performed based on the following equation. That is, l v,4+ = (y hz + V V2 ) (1
/21 On the other hand, the vector length calculator behaves exactly the same as 24. The vector length calculator 25 is Cx, c shown in FIG.
Given y as an input, calculate the absolute value IVAI.

This IVAI becomes the absolute value of the composite vector of all vectors. These absolute values are given to the divider 26,
Here, an operation is performed using IVAI as the dividend and IVMI as the divisor, and the result is output as P.

Through such processing, the value of P is reduced to 0 by the two original data that should determine the hue, in other words, the data of the two dominant channels among R, G, and B, and the value of the remaining one data. It will change in the range of ~100χ. This is because the value of IVAI changes from 0 to the maximum IV, 41, depending on the mixing ratio of the three source data.

As described above, the RGB-P converter 4 (
Fig. 1) converts the degree of saturation (or color purity) determined by three-dimensional original data R, G, and B into one-dimensional data P, making subsequent handling convenient.

Here, the particularities in the field of application of the present invention, namely:
As already explained, if the absolute precision of the converted P value is not important, but only the continuity and monotonically increasing property are important, and the result is not 0~100χ in all cases, the applicant By providing another RGB-P conversion means, it is possible to realize RGB-P conversion with a simpler configuration and without any practical problems.

FIG. 6 shows a specific example of an RGB→P converter different from that shown in FIG.

The vector length calculator and divider indicated by numerals 31 and 32 in FIG. 6, respectively, perform exactly the same operations as the vector length calculator 25 and divider 26 in FIG. 5, respectively. The difference from the configuration in FIG. 5 is that this converter has R and G in FIG.

The point is that an adder 30 is simply arranged instead of a series of circuits from the B input to IVMI.

With this configuration, the output P always varies from O to O depending on the hue.
Although it is not necessarily possible to obtain a result of 10oz,
The most important continuity and monotony are guaranteed. The situation at this time can be explained as follows.

That is, if the original data has an element in only one channel, it is clear that P represents 100χ. On the other hand, for example, in the case of magenta color in which R and B are mixed in equal amounts, IVMI is 2X (IVAI) and P is 0.5. In this way, although the method of FIG. 6 cannot obtain conversion results in the range of O to 100.chi. in all cases, it can function economically and satisfactorily in some cases.

FIG. 7 is a diagram showing the hue and the resulting maximum saturation when the configuration shown in FIG. 6 is adopted.

According to the present invention configured as described above, in the original image expressed by the signals R2O and B, the area belonging to a specific color is assigned to each color attribute independently, that is, the specification of one type of attribute is changed to another attribute. It will be understood that it is possible to specify and extract without affecting it.

DESCRIPTION OF THE EMBODIMENTS FIG. 8 shows an embodiment of an RGB-H converter implementing the functions of the present invention. To explain in detail with reference to the same figure, 4
1a, 41b, and 41C are full adders each having a number of bits equal to the data width of each R, G, and B channel;
B is each input, Σ is output, Ci is carry input, C
O indicates carry output. The output of the full adder 41a is taken out by shifting the upper digit by 1 bit, that is, the original CO is connected as the MSB and the LSB is discarded. As a result, the input to the inverter 42a is effectively (G+
B) Perform the calculation of /2. - The carry input of the labor adder 41b is fixed at logic level 1, and the input to the inverter 42
Since it is given through a, the result is the calculation Σ=B−A. This means that the output of the full adder 41b corresponds to Cx in FIG. That is, G X (-cos 60°) + BX (-CO5
60”) + R= GX (-0,5)+ BX (
-0,5)+ R= (G + B)X O,5+
This is nothing but a violation of the arithmetic formula R.

Further, the full adder 41C receives blue (B) data through the inverter 42b at its B input, and its carry input is fixed at logic level 1, so it performs the operation G-B.

In this way, the output of the full adder 41b represents cy in FIG. 4, and the output of the full adder 41C represents Cy×(1/sin
60°), and each is used as the basis for calculating the arctangent.

In this example, a ROM (Rea) is used to calculate the arctangent.
dOnly Memory) 43 is used. Therefore, the operation of multiplying cy by 5in60'' is R
Since this is done automatically using the OM table, it is advantageous that a special circuit can be omitted.

The ROM 43 is provided with S x % S y terminals for applying the signs of Cx and Cy equivalent signals, and these terminals have 4 terminals.
The carry signals of 1b and 41C are applied.

It is also assumed that the ROM 43 has a terminal H as its output, and the inverse arctangent for 360° is written in advance. By the way, if X and Y are each 7 bits, and the angle output is expressed in 8 bits, the human power will be 16 bits in total, and the required ROM size will be 64 bytes, which is enough for one IC today. It is practical because it can be covered by

Next, FIG. 9 shows the RGB-P converter according to the present invention, that is, the embodiment of FIG. 5. To explain this FIG. 10, 51a, 51b, and 51C each indicate a comparator, and here, the magnitude of the human power A1B is compared, and A>B
The output C is considered to be "true" only when . The combinational logic indicated by the reference numeral 52 is 51a, 51b, 5
Depending on the state of 1C, data of only the largest one and the next largest one among R, G, and B is led to the vector length calculator 53, and other data are inhibited by an AND gate. Here, the truth table for determining the operation of the combinational logic 52 is as shown in Table 2.

Table 2 Since such an operation can be easily realized using a common logic element or a programmable logic element, the internal details will be omitted.

Now, the vector length calculators 53 and 54 have exactly the same structure, and the specific structure is RGB-H shown in FIG.
You can think of it as exactly the same as a converter.

However, the contents of the portion corresponding to the ROM 43 in FIG. 8 are changed to an absolute value calculation table instead of an arctangent calculation table. That is, it is assumed that a comparison table equivalent to the calculation ((SX-X)" + (Sy-Y)")"" has been written. The result is then output and applied to the M and A inputs of the division ROM 55. The ROM 55 has an internal table of P=A/
Since a comparison table equivalent to performing the calculation M is written, the output P expresses one-dimensional data representative of the degree of saturation according to the principle already explained.

Next, based on the principle shown in FIG. 6, an embodiment is shown in FIG. 1st
As shown in FIG. 0, original data R, G, and B are all added together by passing through an adder 61 and an adder 62. As a result, the number of bits of data increases, and the subsequent division R
In order to prevent this, the normalization R
The OM63 adjusts the maximum value of the data to the same number of bits as the original data. The configuration and operation of the vector length calculator 64 and the division ROM 65 are exactly the same as in FIG. 9, so their explanation will be omitted.

Although the method shown in FIG. 11 has the slight drawbacks described above, it is extremely practical and useful due to its simple configuration.

Next, Fig. 11 shows an example of a window comparator for extracting a specific area related to the attributes of ■ and P, but this point is an application of the prior art and will not be explained as it does not require any special explanation. Omitted.

On the other hand, for the H attribute, the numerical values form a ring, that is, the hue with the maximum numerical value is connected to the hue with the minimum numerical value, so
It is necessary to create a window comparator. Such a window comparator is described in the specification of a device titled "Window Discriminator for Data Having a Circulating Numerical Code," which was filed by the applicant as a utility model on August 31, 1985. It is preferable that

Next, an embodiment of the AND circuit 6, which is the final stage of the basic configuration shown in FIG. 1, will be explained using FIG. 12.
In principle, the simple AND circuit 6 shown in FIG. 1 is sufficient for its purpose.

This is clear from the explanation in section 4) on page 9 of this specification. However, depending on how the present invention is applied, it may be desirable to intentionally ignore any one or two of the three attributes (l H, P). For example, when it is desired to perform region extraction when R, G, and B are mixed and captured as one monochrome image, the existence of attributes other than ■ becomes an obstacle. There are also cases where it is desired to ignore brightness and saturation and simply consider hue. The configuration shown in FIG. 12 functions well as an AND circuit with such expandability.

Now, in FIG. 12, ID, HD, and PD are signals that are recognized as specific areas through the window comparator, and IM, 1-fM, and PM are signals that are recognized as specific areas through the window comparator, and IM, 1-fM, and PM are signals that are associated with ID, HD, and PD ( Ignore) signal. Here, if the override signal is "true"
When , the area signal related to this is unconditionally “true”
becomes. This behavior is equivalent to the fact that there is no corresponding path in the final AND gate. Therefore, Table 3 summarizes how the device performs recognition depending on the states of IM, HM, and PM.

Table 3 Effects of the Invention As described above, by using the method of the present invention, it is possible to extract a specific color region based essentially (not equivalently) on the three attributes of color in the fields of image processing and image analysis. Moreover, the settings of each attribute parameter can all be specified using simple one-dimensional data, and furthermore, any attribute parameter can be changed independently without affecting other parameters. Furthermore, since it is possible to freely select the attributes of interest, it is possible to demonstrate an extremely practical function.

In the above description, reference was made to a color image consisting of three channels as the original data, but the present invention can of course be applied to a multiband image consisting of any number of channels greater than or equal to two channels. Further, in the above description, only the extraction of a region in an image has been described, but it is clear that the same configuration can similarly identify the color of a specific region of an original image.

[Brief explanation of drawings]

FIG. 1 is a circuit diagram showing the basic configuration for implementing the method of the present invention. FIG. 2 is an explanatory diagram showing a state in which a desired color on a hue wheel is expressed by an angle. FIG. 3 is an explanatory diagram showing a hexagon drawn by a composite vector when the original data R, G, and B are considered as vectors. FIG. 4 is a circuit diagram illustrating the configuration of an RGB→H converter, which is one of the features of the present invention. FIG. 5 is a circuit diagram illustrating the configuration of an RGB-P converter, which is one of the features of the present invention. FIG. 6 is a circuit diagram explaining another configuration of the RGB4P converter. FIG. 7 is an explanatory diagram showing the hue and the maximum saturation degree obtained by the hue. FIG. 8 is a circuit diagram showing an embodiment of the RG B-H converter. FIG. 9 is a circuit diagram showing an embodiment of the RGB-P converter related to FIG. 5. FIG. 1O is a circuit diagram showing an embodiment of the RGB-H converter related to FIG. 8. FIG. 11 is a circuit diagram showing an embodiment of a region extractor using I and P attributes. FIG. 12 is a circuit diagram showing an embodiment of a practical AND circuit. 1... Original signal generator 2... ROB→I converter 3... RGB-H converter 4... RGB-4P converter 5 i, 5h, 5p... Window comparator 6...
・Logic product circuit 7...Output 21...Maximum value selector 22...Vector sum calculator 23...Next maximum value selector 24.25...Vector length calculator 26...Divider 30...Adder 31...Vector length calculator 32...Divider 41a, 41b, 41cm-Full adder 42a, 42b...Inverter 43...ROM for arctangent calculation 51a, 51b, 51C...Comparator 52...Combination logic 53.54...Vector length calculator 55...Division ROM 61.62...Adder 63...Normalization ROM 64...Vector length calculator 65 ...Division ROM

Claims (6)

[Claims] (1) In order to extract or identify a specific color from a multiband image recorded in at least two or more channels, the level information for each channel represents the color attributes (brightness, hue, saturation). Convert to vector information,
This is a method of discrimination, in which the sum of the absolute values of each band multiplied by an appropriate coefficient is used as a means of converting to lightness axis components, and furthermore, the sum of the absolute values of each band is multiplied by an appropriate coefficient, and further, as a means of converting to hue axis components, the sum of the absolute values of each band is used. When calculating this angle, each vector representing each channel is made with the reference axis of the orthogonal coordinates. A coefficient unit that calculates the cosine component of the angle formed by the angle, a coefficient unit that calculates the sine component, and an adder that calculates the sum of each component are used to calculate the cosine component of the composite vector, the arctangent that takes the sine component as input, and the output of the calculator A specific color region extraction method characterized by using. (2) The specification according to claim 1, characterized in that when identifying a specific expression area by adding each color attribute quantity to a window comparator and using the ethical product of the comparator output, the comparator output related to each color attribute is masked. Color region extraction method. (3) In order to extract or identify a specific color from a multiband image recorded in at least two or more channels, the level information for each channel represents the color attributes (brightness, hue, saturation). Convert to vector information,
This is a method for performing discrimination, in which the sum of the absolute values of each band multiplied by an appropriate coefficient is used as a means of changing to the lightness axis component, and the above-mentioned method is used as a means of changing to the saturation axis component. The absolute value of the composite vector is |V_H|, and the largest vector 2 represents the level of each channel recording the original image.
A method for extracting a specific color area, characterized in that the absolute value of a composite belt having only one element as |V_F| is used, and the output of an arithmetic unit that calculates |V_H|/|V_P| is used. (4) The specification according to claim 3, characterized in that when identifying a specific expression region by adding each color attribute amount to a window comparator and using the ethical product of the comparator output, the comparator output related to each color attribute is masked. Color region extraction method. (5) In order to extract or identify a specific color from a multiband image recorded in at least two or more channels, the level information for each channel represents the color attributes (brightness, hue, saturation). Convert to vector information,
This is a method for performing discrimination, in which the sum of the absolute values of each band multiplied by an appropriate coefficient is used as a means of changing to the lightness axis component, and the above-mentioned method is used as a means of changing to the saturation axis component. The absolute value of the composite belt is |V_H|, and the sum or average of the levels for each channel recording the original image is L, |V_H|
A specific color area extraction method using the output of a computing unit that performs the calculation /L. (6) The specification according to claim 5, characterized in that when identifying a specific expression area by adding each color attribute amount to a window comparator and using the ethical product of the comparator output, the comparator output related to each color attribute is masked. Color region extraction method.