JP3182112B2 - Building exterior color determination support system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a support system for determining the color of the exterior of a building, and more particularly to a support system capable of giving information about the color when determining the color of the outer wall of the building. .

[0002]

BACKGROUND OF THE INVENTION It is known that the exterior walls of buildings have a significant effect on the overall street impression evaluation. In particular, in the case of a mid- to high-rise office building with a large occupied area, the influence is large. As described above, the determination of the color is one of the important factors in the design of a building, but at present, the determination of the color depends on the creativity and sensibility of the designer.

On the other hand, from the study of colors, the image of each color (the emotion of a person who has seen a certain color) has been found to some extent based on statistics. However, the image that the color itself has and the image generated as the color of the outer wall formed by painting it on the outer wall of a large area do not always match, and the research results of such a color cannot be applied as it is.

Conventionally, in order to quantify the color image of the exterior wall of a building, a mathematical model formula is assumed as a hypothesis for some colors, and a multivariate analysis (multiple regression analysis, quantification,
In some cases, the hypothesis is verified by a covariance analysis and the model formula is used as a quantification model. However, such a method requires trial and error to determine the model formula of the hypothesis, which is complicated. Moreover, there are infinite types of colors including intermediate colors, and it is impossible to determine a model formula for all such infinite colors. Therefore, such a method cannot be put to practical use.

Therefore, the present inventor, when determining a color (physical quantity) that exerts a certain image (psychological quantity),
We considered using a neural network. That is, as the input information, the SD method (semantic d) is used.
Sixteen sets of adjective pairs (SD scale evaluation values) according to the differential method were used. As is well known, the SD method is a set of adjective pairs (such as "bright-dark") representing the opposite meaning as shown in FIG. In order to evaluate the image, 16 sets of adjective pairs used in evaluating the color were picked up and used.

As a neural network, a typical input layer (the number of cells corresponding to an adjective pair is 16
), And a three-layer structure composed of an intermediate layer and an output layer. The output layer has three colors, “hue”, “saturation”, and “brightness” for specifying colors. Then unknown (undefined)
Although there is an advantage that can be quantified, the hues were slightly better in terms of hue than those using the conventional model formula, but the saturation and lightness were the same as before. When the difference between the color determined by using a single neural network and the actual value (teacher value) is verified, it is better than the conventional one, but there is still no one that can be put to practical use. Was.

Accordingly, the present inventors have proposed a method as shown in FIG.
We considered building a hierarchical neural network. Here, the neural networks of the second hierarchy are arranged in parallel according to N groups. Each neural network has the same configuration as the above-mentioned single neural network, and has 16 sets of adjectives. A pair (SD scale evaluation value) is input, and three elements for specifying a color are determined and output. The first-layer neural network is a network that inputs an SD scale evaluation value and outputs which network of the second layer is suitable for recognizing colors. Therefore, the number of output cells is equal to the number (N) of the number of neural networks in the second hierarchy.

In this two-layer network,
First, it is determined to which group the color corresponding to the image specified by the SD scale evaluation value input in the first hierarchy belongs. More specifically, the degree of suitability / reliability indicating an index of the possibility that a color that matches the image belongs to each group is output. Next, the same S as the one input to the first hierarchy
The D scale evaluation value is input to each network of the second hierarchy,
At the second level, one of the colors belonging to its own group is specified and output. Then, the determined color information and the reliability are output as a pair. As a result, the colors output from the group having the highest reliability are the ones most suitable for the image.

The groups are classified by performing cluster analysis based on the image information of each color to cluster those having a high correlation. As a result, 100 colors were able to be divided into five groups, and specifically, N was 5 as described above.

By dividing the network into groups as described above, the learning space of each network is narrowed, the learning effect is improved, and the recognition accuracy is also improved. As a result, although the recognition accuracy was improved as compared with the prior art quantified (color estimation) using the above-described single neural network, sufficient accuracy was not obtained for practical use. The technique of estimating the color corresponding to the image (SD scale evaluation value) using the single and two-layer neural networks described above is a prior art proposed when creating the present invention, and is not a known technique. .

SUMMARY OF THE INVENTION The present invention has been made in view of the above background, and has as its object to solve the above-described problems and to estimate and determine the color (candidate) of an outer wall of a building that matches an image. The present invention provides a color determination support system for a building external appearance, which can make the input operation easier. In addition, it is possible to accurately estimate the image information that is exhibited when a certain color is used as the color of the outer wall, and to check the image of the color determined by the designer, etc. Another purpose is to provide a system.

[0012]

In order to achieve the above-mentioned object, in a system for supporting the determination of the color of the exterior of a building according to the present invention, an image profile consisting of SD scale evaluation values of a predetermined number of adjective pairs is inputted, and A building exterior color determination support system for estimating and outputting a color candidate of a building outer wall corresponding to an image profile, comprising: a reliability calculating unit including a neural network for calculating a reliability for each hue classification; Brightness / saturation determining means provided in parallel for each of the above, hue determining means for receiving an output of the brightness / saturation determining means and determining a hue, and a predetermined color candidate obtained by each of the means. Output means for outputting the brightness, saturation, hue, and reliability in association with each other. The lightness / saturation determination means includes a neural network that uses the image profile as input data, estimates and outputs the lightness and saturation of the hue classification. Further, the hue determination means is configured to include a function of estimating information on hue by a neural network using the image profile and the lightness and saturation output from the corresponding lightness / saturation determination means as input data. (Claim 1).

That is, according to the present invention, the reliability of each hue segment is first obtained by the reliability calculating means. Therefore, the lightness / saturation determining means for each hue category and the reliability of the color candidate determined by the hue determining means are determined. You can know. As a result, it is possible to know how probable it is, so that it is possible to accurately determine whether or not the output color candidate is finally determined as the color of the outer wall.

Further, since each of the determination units is configured using a neural network, it can cope with input of unknown data. In addition, since the neural network is formed by dividing each hue segment, the learning area is narrowed, the learning can be performed efficiently and accurately, and a good learning result can be obtained. In addition, the brightness and the saturation are first determined, and the hue can be determined more accurately because the brightness and the saturation determined earlier are used as input data together with the image profile when the hue is determined.

Further, the neural network in the hue determination means includes hue parameters MA and M shown below.
MA = saturation × cos (hue angle) MB = saturation × sin (hue angle) The apparatus further comprises means for calculating a hue based on the outputs MA and MB of the neural network. (Claim 2). In this way, when the hue angles are compared with each other, for example, the hue angles are 0 degree and 355 degrees, the similar hue can be obtained by using the hue parameter even when the hue angles are apart from each other but are actually similar. Is closer to the value of the hue parameter and more accurate.

An image profile determining means for creating an image profile to be input to the system according to claim 1 or 2, further comprising:
Accessing a storage unit storing a template in which an image relating to a color prepared in advance and an SD scale evaluation value for each adjective pair corresponding to the image are stored, displaying predetermined data on a display unit, and using the template. It may be configured to have a function of determining the image profile (claim 3). In this way, even if the number of adjective pairs increases, an image profile can be created relatively easily and accurately.

When each of the above-mentioned systems is operated, the lightness / saturation determining means and the hue determining means for all hue categories are operated regardless of the result of calculation by the reliability calculating means. (Claim 4). In this case, the operation algorithm becomes constant, so that control becomes easy.

On the other hand, the lightness / saturation determining means and the hue determining means may be operated for a hue classification in which the reliability calculated by the reliability calculating means satisfies a predetermined criterion. In this way, color candidates are actually determined only for those with high reliability, so that unnecessary calculation processing is not executed.

The color candidates displayed on the output means may be a plurality of colors determined based on the reliability calculated by the reliability calculation means. That is, in the present invention, since the brightness / saturation determining means and the hue determining means each including a neural network are provided for each hue section, the color candidate determined by each means is one.
Even if it is one, it is possible to obtain the color candidates as many as the maximum number of hue sections as a whole (of course, in the case of claim 5, the number of actually operating hue sections). Therefore, according to the fifth aspect, when one image is input, a plurality of color candidates can be output, and the range of selection can be expanded. Note that “based on the reliability” refers to, for example, a relatively selected value such as the top n reliability values (five in the embodiment) or a reliability value equal to or higher than a certain reference value. It is possible to adopt various methods such as an absolute selection, such as performing the above, or a combination thereof.

[0020]

[0021]

The neural network constituting the color determination support system is provided with a plurality of types of sample images in which a predetermined color is actually applied to the appearance of a building, and each sample image is shown to a plurality of persons. It can be constructed by learning what obtained the SD scale evaluation value of the predetermined number of adjective pairs for the image as teacher data (claim 7).

That is, various studies have been made on the SD scale evaluation value of the adjective pair for the color itself. However, when a color is applied to the appearance of a building, it often differs from the SD scale evaluation value of the color itself. Therefore, we prepared several kinds of sample pictures with various colors actually painted on the exterior of the building, and when we showed it to multiple people, we heard the SD scale evaluation value from each person,
Based on this, an SD scale evaluation value for the color of the building is obtained to create teacher data. By learning the neural network using the teacher data, it is possible to accurately determine the association between the colors other than the sample images and the image profiles. Note that many types of sample images can be easily prepared by using computer graphics or the like.

[0024]

FIG. 3 shows a preferred embodiment of a building exterior color determination support system according to the present invention. In this embodiment, an image profile for a building outer wall possessed by a user such as a designer is input based on SD scale evaluation values for 30 sets of adjective pairs, and 3
Finally, hue candidates are determined and output using a neural network having a hierarchy.

That is, as shown in FIG. 2, various input data are given to the image profile determining unit 2 using an input device 1 such as a keyboard or a mouse, and predetermined data is sent from the image database 4 as needed. Read out and create an image profile of the colors you need.

First, a specific example of an image profile serving as input data will be described. An SD scale evaluation value composed of 30 pairs of adjectives shown in FIG. 4 is used. And this 3
As shown in the figure, the evaluation factor (eval)
), potential factors (potency), and activity factors (activity), so that each factor can be input.

As an input method, first, an image profile input screen as shown in FIG.
To be displayed. The input screen shown in FIG. 5 is a master screen, above which is a factor selection key K1, and below it is a display area R for displaying adjective pairs belonging to the selected factor.

Then, the input device 1 such as a mouse is operated to move the pointer P to a display area of the factor to be inputted, and by clicking in that state, the selected factor (the active factor in the illustrated example) is selected. A list of belonging adjective pairs is displayed in the lower display area R. In this display area R, an evaluation value input key K2 for inputting an SD scale evaluation value for each adjective pair is displayed, and by moving the evaluation value input key K2, the scale position corresponding to the image is displayed. The evaluation value for each adjective pair is determined by moving the key K2 one by one to the clicked side by moving or clicking on a blank portion existing on both sides of the key K2.

By the way, it is troublesome to input the SD scale evaluation value for each of the 30 adjective pairs while imagining the input as described above. Further, for example, an image in which the evaluation value of “elegant” is the maximum (−3) and the evaluation value of “sloppy” is the maximum (3) is generally contradictory. As described above, it is complicated to input evaluation values of a pair of adjectives without contradiction.

Therefore, in this embodiment, a template in which a plurality of representative images are associated in advance with the SD scale evaluation values of each adjective pair is prepared, and the SD scale evaluation values of 30 sets of adjective pairs are determined using the template. I can do it. Specifically, the “template” portion K3 displayed on the master screen shown in FIG. 5 is selected. Then, a template input screen as shown in FIG. 6 is displayed. Although both the template input screen and the master screen are displayed in a multi-window, the display may be switched to the template input screen.

Then, a template matching the image is selected from the displayed template candidates (“template name”).
Is displayed in the column of “Apply”, and if that is sufficient, the “application” part K4 is designated. As a result, the SD scale evaluation value of each adjective pair associated with the template is read, and each key K in the display area R of the master screen shown in FIG.
The position 2 moves to a position corresponding to the read SD scale evaluation value. FIGS. 7 to 9 show examples of the SD scale evaluation value of each adjective pair associated with the three templates shown in FIG.

The SD scale evaluation value stored as the displayed template may be used as it is, or a predetermined adjective pair may be appropriately corrected based on the SD scale evaluation value. Also, even if the correction is made, the target image profile can be created more easily and in a shorter time than in the case where all settings are made from the beginning because of the model. in this way,
By using the template, the SD scale evaluation value corresponding to the image assumed by the user can be easily determined, which is convenient.

As shown in FIG. 6, by appropriately using "add", "delete", and "rename" buttons,
Maintenance such as adding a new image profile template or deleting unnecessary templates can be performed. As described above, the master screen, the template input screen of the image profile, and the SD scale evaluation value data of each adjective pair associated with the template are stored in the image database 4.
, And appropriate data is read out according to the input data and displayed on the display device 5, or predetermined data is stored in the image database 4.

When the SD scale evaluation values for all three factors are determined in this way, the determined SD
The scale evaluation value is provided to the reliability calculation section 8 of the first hierarchy in the color candidate calculation section 7 and the brightness / saturation determination section 9 of the second hierarchy. The specific internal configuration of the color candidate calculation unit 7 is as shown in FIG.

First, the reliability calculating section 8 of the first hierarchy calculates the reliability for each hue based on the given image profile (SD scale evaluation value).
That is, in the case of the prior invention using the two-layered neural network described in the background section of the invention, the reliability of each of the five groups obtained as a result of clustering in the first layer is calculated. However, in the present invention, grouping is performed for each hue (hue classification), and the degree of conformity (reliability) for each hue is determined.

That is, the hue classification is “RP, R, YR,
Y, GY, G, BG, B, PB, P ”, and each hue classification can be further classified into a predetermined number of stages (1 to 10 RP if RP). The hue classification can be represented by dividing 360 degrees sufficiently at equal intervals (36 degrees) as shown in FIG. 11 in the musical coordinate system. Therefore, using the neural network A, the degree of suitability / reliability indicating an index of the possibility that a color suitable for the input data (SD scale evaluation value) belongs to each hue classification is calculated and output. Since the reliability determined here is output as it is, the hue classification having the highest reliability is a hue candidate having the highest possibility that the color corresponding to the input image belongs to. The neural network A used in the reliability calculation unit 8 has an input layer composed of 30 cells (corresponding to the number of adjective pairs), a predetermined number of intermediate layers, and 10 cells (the number of hue classifications). A general configuration having an output layer having (correspondence) is used.

Further, in the musical coordinate system shown in FIG. 11, the hue is specified as described above, but the saturation increases as the distance from the center increases. Further, the brightness in the Z-axis direction of the curved coordinate system shown in FIG. In other words, each color exists in any of the three-dimensional spaces formed by stacking the circles shown in FIG. 11 in an amount corresponding to the number of lightnesses, and by defining the colors, It is specified whether or not it exists at the position. The distance from the center is determined by defining the saturation, and the position in the Z-axis direction (height direction) is determined by defining the lightness. Thereby, one color is specified.

Therefore, in the brightness / saturation determination section 9 which is the second layer, ten pieces are provided in parallel corresponding to the ten kinds of hue divisions, and each 1
The same SD scale evaluation value as that input to the hierarchy is input to the neural network B, and each lightness / saturation determination unit 9 specifies the lightness V and the saturation C in the colors belonging to its own hue, Output to the hue determination unit 10 for the corresponding hue at the next stage. The neural network B used in the brightness / saturation determination unit 9 includes an input layer including 30 cells (corresponding to the number of adjective pairs), a predetermined number of intermediate layers, and two cells (brightness and saturation). A general configuration having an output layer composed of the above-mentioned components is used. Note that the reliability for each hue classification calculated by the reliability calculation unit 8 in the first hierarchy is not reflected in the determination of lightness / saturation and is passed through as it is.

In this embodiment, in order to standardize the processing, the lightness / saturation determination units 9 for all hues are operated irrespective of the magnitude of the reliability calculated by the reliability calculation unit 8. The brightness and saturation are determined and output respectively. Therefore, for example, the reliability calculation unit 8 and the brightness / saturation determination unit 9 may be operated in parallel.

However, in the present invention, it is not always necessary to operate the lightness / saturation determination unit 9 for all hues as described above, and the hue selected based on a certain reference based on the result of the reliability calculation unit 8. Only the brightness / saturation determination unit 9 may be operated. The certain reference in that case may be a relative one such as the top N in descending order of reliability, or an absolute one such that the reliability is X or more.

The hue is finally determined by the hue determination section 10 which is the third layer. The hue mentioned here is not the 10 kinds of hue divisions obtained by the reliability calculation unit but the minimum unit obtained by further dividing each hue division into a predetermined number. That is, 1RP-1
0RP. Further, in the present embodiment, the output of the hue determination unit 10 for a certain hue is not always the hue belonging to the hue category, and there is a possibility that a hue belonging to another hue category is selected. As an example, for example, the hue R
Is calculated and determined by the hue determination unit 10 for mR (m
Is not a predetermined number from 1 to 10), but may be, for example, 7RP.

As shown in FIG. 10, the internal structure of each hue determination unit 10 is the brightness V and saturation C determined by the corresponding brightness / saturation determination unit 9 and the image profile (corresponding brightness / saturation determination). Neural network C, which receives three types of input data, which are provided through the unit 8 or directly from the image profile determination unit 2).
, And two parameters (M
A, MB) is calculated and output. Further, based on the parameters (MA, MB), the hue H
To determine. The hue H determined by the hue determination section 10 is output to the output control section 11 in association with the brightness V and the saturation C determined by the brightness / saturation determination section 9 and the reliability calculated by the reliability calculation section 8. I will be.

Here, the parameters (MA, MB) for specifying the hue H will be described. Simplification of the music coordinate system shown in FIG. 11 can be represented as shown in FIG. If 10RP is set to 0 degree, the hue of a certain color Q can be specified by the hue angle θ, and the saturation is also specified by the distance C from the center. In other words, (θ,
In C), the coordinates of Q can be specified. By doing so, for example, 10RP with a hue angle θ of 0 degrees and 5RP with a hue angle θ of 342 degrees
In comparison with R, the color is close, but 342 when compared by the hue angle.
Very different hues are separated from each other. Therefore, in the present embodiment, a graph of music coordinates defining colors is shown in FIG.
2, the hue angle of 0 degree is taken as the X axis, and the position of the hue angle of 90 degrees is taken as the Y axis. And
The parameters MA and MB are as follows: MA = C × cos θ MB = C × sin θ where C is a saturation and θ is a value defined by a hue angle. That is, among the coordinate values of a certain color Q in the rectangular coordinate system, X
The coordinate value of the axis becomes MA and the coordinate value of the Y axis becomes MB. Thus, when the saturation C is the same and the hue angles are 0 degree and 342 degrees, the former becomes (C, 0) and the latter becomes (0.95).
C, -0.31C), and it can be determined that the hues are relatively close.

The learning of the neural networks of the second and third layers is performed in a narrow range of the same hue division, so that the efficiency and accuracy are improved, and the same sample is used for learning. Since the hue classification and the hue classification adjacent to the hue classification are performed together, the number of samples increases, and more accurate learning can be performed. As a result, the learning result becomes good, and a good recognition result can be obtained even in actual use.

The output control section 11 outputs to the display device 5 in accordance with a predetermined display format based on the color candidate data provided from the color candidate calculation section 7. As shown in the display example, as shown in FIG. 13, the top five items are displayed in descending order of reliability. The numbers from 1 to 5 at the upper left of the area displaying each color candidate are candidate ranks with high reliability. An area candidate R1 for displaying reliability is provided at the upper left of each area, and the right half is provided at the right half. An area R2 is provided in which three elements (hue H, lightness V, and saturation C) for specifying a color are numerically displayed. Each area R
1, the value displayed in R2 uses the value output from the color candidate calculation unit 7 (however, since the reliability takes a value of 0 to 1, the value is multiplied by 100 and converted to%, and the conversion is performed. Rounded off after the decimal point).

Further, if the three elements are simply indicated by numerical values, it is difficult for a viewer to recognize the actual color. Therefore, in this embodiment, the color conversion unit 12 is provided, and the three elements (H, V,
The color specified in C) is converted into RGB data, and the color corresponding to the color sample display region R3 is displayed based on the converted color data. This allows the user to display the colors that match the user's image in order from the one with the highest degree of reliability, so that the user can use the colors effectively as information when finally determining the color of the building outer wall. The output control unit 11 can also print out the data on the extracted color candidates via the printer 13 as necessary.

The hue obtained when the finally determined lightness / saturation / hue is obtained when the reliability (0 to 1) obtained by the reliability calculation unit 8 becomes 0.3 or more. FIG. 14 shows the relationship between the angle and the actually measured value. Those with a reliability of 0.9 or more were circled.
As is clear from the figure, the difference between the actual measurement value and the calculation result decreases as the reliability increases, and the difference between 0.9 and 0.9 almost agrees. Therefore, even if a color candidate is determined for an unknown input, its reliability and certainty are guaranteed.

FIG. 15 shows a second embodiment of the color determination support system according to the present invention. As shown in the figure, in the present embodiment, contrary to the above-described first embodiment,
When the color of the exterior wall of the building is determined, it predicts what the image of the color will be and displays it, so that the image is as expected or close to the image that the designer assumed. It is used as a material for verifying whether or not the color is finally determined.

More specifically, as shown in the drawing, the input device 1 such as a mouse or a keyboard is used to provide various information to the color information determination unit 20. Display device 5
The color information is determined while displaying the information. As an example, the three elements of color (H,
V, C), an input screen as shown in FIG. The three elements of hue, lightness, and saturation may be directly input using the input device 1, or the color may be determined by selecting a model color displayed below. In this selection method using the model colors, first, a predetermined pattern is selected from the model patterns MP1 for the hue. Then, the upper hue column is quantified based on the numerical value. In addition, since the model pattern MP2 of the brightness and the saturation corresponding to the hue is displayed, the user selects the color assumed by the user. Then, the upper lightness and saturation columns are digitized based on the numerical values.

As another input method, the "set by color name" field on the screen shown in FIG. 16 is selected. Then, a screen displaying a list of color names as shown in FIG. 17 is displayed. The user can also input a color by searching for a color name that specifies the color determined by the user and selecting the color name. As a result, the three elements (HV)
Even if you do not know C), you can input based on a conventional color name,
It is convenient.

As still another input method, FIG.
As shown in (1), a color can be specified by RGB data and input. In this case as well, the screen shown in FIG. 17 can be read out by selecting the “set by color name” field. When specifying by RGB, the RGB
The data is sent to the color conversion unit 22, HVC data indicating a color specified from the RGB data is obtained, and the HVC data is returned to the color information determination unit 20.

Further, regardless of which method is adopted, the color sample selected at that time is displayed at the upper left of the screen, so that it is possible to easily confirm whether or not the color is the one assumed. It has become. Then, in any of the input screens, by selecting the “image prediction” field, the color information specified on the input screen at that time is transferred to the image prediction unit 24. As the data to be transferred to the image prediction unit 24, the lightness and the saturation may be the same, but for the hue, the above two parameters (MA, MB) are obtained and given. That is, since the hue angle is known from the hue, the parameters MA and MB can be obtained based on the hue angle and the saturation C.

The image prediction unit 24 uses a plurality of 4-input and 1-output neural networks, and obtains an SD scale evaluation value for one adjective pair in each neural network. Accordingly, in this embodiment, since 30 adjective pairs are used, the neural network is also 3
0 units are installed in parallel.

Then, SD profile evaluation values of each adjective pair output from each neural network are put together to create image profile data, and transferred to the output control unit 25. The output control unit 25 classifies the given image profile data for each of the three factors.
As shown in FIG. 7, the SD scale evaluation values of a predetermined number of adjective pairs (five pairs in the illustrated example) are output by classifying each factor. Then, by operating the scroll bar via the input device 1, it is possible to display the SD scale evaluation value of another adjective pair belonging to the same factor. Further, by switching the column of the factor, the SD scale evaluation value of the adjective pair belonging to another factor can be displayed. Thereby, the SD scale evaluation values of all the adjective pairs can be confirmed on the screen of the display device 5. Further, it is also possible to print out all or a part using the printer 13.

[0055]

As described above, in the building appearance color determination support system according to the present invention, it is possible to estimate and determine the color (candidate) of the outer wall of the building according to the image by using a three-layer neural network. become able to. Further, when configured as defined in claim 3, the input operation can be facilitated.

Further, in the case of the configuration defined in claims 7 and 8, image information exhibited when a certain color is used as the color of the outer wall can be accurately estimated.
It is possible to confirm the image of the color determined by the designer.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating adjective pairs by the SD method.

FIG. 2 is a diagram illustrating a prior invention (not known) developed by the present inventors until the present invention was devised.

FIG. 3 is a block diagram showing a first embodiment of a building exterior color determination support system according to the present invention.

FIG. 4 is a diagram illustrating SD scale evaluation values of adjective pairs used as input data.

FIG. 5 is a diagram illustrating a function of an image profile determination unit.

FIG. 6 is a diagram illustrating a function of an image profile determination unit.

FIG. 7 is a diagram illustrating a function of an image profile determination unit.

FIG. 8 is a diagram illustrating a function of an image profile determination unit.

FIG. 9 is a diagram illustrating a function of an image profile determination unit.

FIG. 10 is a diagram showing an internal configuration of a color candidate calculation unit.

FIG. 11 is a diagram illustrating hue.

FIG. 12 is a diagram illustrating hue parameters (MA, MB).

FIG. 13 is a diagram illustrating functions of an output control unit.

FIG. 14 is a diagram for demonstrating the effect of the first embodiment.

FIG. 15 is a block diagram showing a second embodiment of a building exterior color determination support system according to the present invention.

FIG. 16 is a diagram illustrating a function of a color information determination unit.

FIG. 17 is a diagram illustrating a function of a color information determination unit.

FIG. 18 is a diagram illustrating a function of a color information determination unit.

FIG. 19 is a diagram illustrating a function of a color information determination unit.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input device 2 Image profile determination part 4 Image database 5 Display device 7 Color candidate calculation part 8 Reliable part calculation part 9 Lightness / chroma determination part 10 Hue determination part 11 Output control part 12 Color conversion part 13 Printer 20 Color information determination part 21 database 22 color conversion unit 24 image prediction unit 25 output control unit

──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Tosumi Sawada 4-38-13 Honcho, Nakano-ku, Tokyo Stock Company Structure Research Institute (72) Inventor Daisuke Yamamoto 4-38-13, Honcho Nakano-ku, Tokyo Stock (56) References JP-A-64-88320 (JP, A) JP-A-2-170281 (JP, A) JP-A-9-6573 (JP, A) JP-A-8-50648 ( JP, A) JP-A-6-203118 (JP, A) JP-B-56-46082 (JP, B2) Building exterior wall color selection system using neural network. Quantification of Building Exterior Color Image Hitoto Sato Kazumi Nakayama Ergonomics Vol. 33 No. Supplement P.S. 322-323 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06F 17/50 680 G06F 17/50 604 Patent file (PATOLIS) JICST file (JOIS)

Claims (7)

(57) [Claims] 1. A color determination support system for a building external appearance, comprising inputting an image profile comprising SD scale evaluation values of a predetermined number of adjective pairs and estimating and outputting a color candidate of a building outer wall corresponding to the image profile. A reliability calculating means having a neural network for calculating the reliability for each hue section; a brightness / saturation determining means provided in parallel for each hue section; and an output of the brightness / saturation determining means. And hue determining means for determining hue, and output means for associating brightness, chroma, hue and reliability of the predetermined color candidate obtained by each of the means, and outputting the lightness / chroma. The determining means includes a neural network that receives the image profile as input data, estimates a brightness and a saturation of a hue classification thereof, and outputs the estimated hue classification. The exterior of the building is characterized in that the determining means includes a function of estimating information on hue by a neural network using the image profile and the lightness and saturation output from the corresponding lightness / saturation determining means as input data. Color decision support system. 2. The neural network in the hue determination means estimates hue parameters MA and MB as follows: MA = saturation × cos (hue angle) MB = saturation × sin (hue angle) 2. The system according to claim 1, further comprising means for calculating a hue based on network outputs MA and MB. 3. An image profile determining means for creating an image profile to be input to the system according to claim 1 or 2, wherein said determining means includes an image relating to a color prepared in advance and each adjective pair corresponding to the image. A storage means storing a template in which an SD scale evaluation value is associated, displaying predetermined data on a display means, and using the template to determine the image profile. Claim 1.
Or the color determination support system for building exterior according to 2. 4. The apparatus according to claim 1, wherein said brightness / saturation determination means and hue determination means are operated for all hue categories irrespective of the result of calculation by said reliability calculation means. A building exterior color determination support system according to any one of the preceding claims. 5. A method according to claim 1, wherein said reliability calculated by said reliability calculation means is a brightness / lightness / height division for a hue segment satisfying a predetermined standard.
The color determination support system for a building exterior according to any one of claims 1 to 3, wherein the saturation determination means and the hue determination means are operated. 6. The color candidate to be displayed on the output means is a plurality of colors determined based on the reliability calculated by the reliability calculation means. The color determination support system for building exteriors described in the section. 7. The neural network that constitutes the color determination support system includes: preparing a plurality of types of sample images in which a predetermined color is actually applied to the appearance of a building; and displaying each sample image to a plurality of persons. The building appearance according to any one of claims 1 to 6, wherein a result of calculating an SD scale evaluation value of the predetermined number of adjective pairs for each sample image is learned as teacher data. Color decision support system.