JP2001195579A - Image evaluating device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To sensitively evaluate an image while using an adjective. SOLUTION: In a step S11, an image feature amount extracting part judges whether an image has been inputted or not and when it is judged that the image has been inputted, in a step S12, the feature amount (hue, lightness, saturation, number of colors and curvature or the like) of the image is extracted and outputted to a feature amount comparing part. In a step S13, while referring to an adjective list by categories registered in an adjective data base, the feature amount comparing part calculates the difference of the feature amount between the inputted image and each of adjectives and selects an adjective having a feature close to the feature of the inputted image on the basis of the calculated result, and a 'weight' expressing the likelihood of that adjective is calculated and outputted to a sentence constitution part. In a step S14, while using the adjective corresponding to the image and the weight, the sentence constitution part constitutes a sentence showing the sensitive evaluation of the image and outputs the constituted sentence in a step S15.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image evaluation device, which extracts features of an input image, compares the features with features corresponding to adjectives registered in a database in advance, and uses the result. The present invention relates to an image evaluation device that can express a sentence using adjectives by expressing a sentence by creating and outputting a sentence.

[0002]

2. Description of the Related Art Conventionally, features of image data are extracted by using a plurality of factors such as lightness, the colors most frequently used, and the distribution of black pixels in a binarized image, and converting them into numerical values. Have been run by. For example, when searching for an image similar to a certain image from an image database in which a plurality of images are stored, a plurality of elements such as brightness, color, and distribution of black pixels of the binarized image are represented by a multidimensional vector. The similarity of each image is calculated by comparing the result of synthesis (scalar amount) of a key image and a multidimensional vector of each image in the database.

[0003]

However, according to the conventional image data feature extraction method based on the measured numerical values, when it is desired to sensibly capture the features of the image data, for example, "cute" or "cool" It is difficult to evaluate and compare image data using such adjective elements.

The present invention has been made in view of such circumstances, and extracts features of an input image, compares the features with features corresponding to adjectives registered in advance in a database, and compares the features. By creating and outputting a sentence using the result, it is possible to express the evaluation of the input image as a sentient sentence using an adjective.

[0005]

According to the present invention, there is provided an image evaluation apparatus comprising: a feature extracting unit for extracting features of an input image;
Holding means for holding an adjective and a feature corresponding to the adjective; calculating means for calculating a difference between a feature of the image extracted by the feature extracting means and a feature corresponding to the adjective held in the holding means; And a control unit for controlling the output of the sentence created by the sentence creation unit using the result of the calculation.

[0006] The feature extraction means is, for example, an image feature quantity extraction unit 41 of FIG.
According to the adjective database 34 in FIG. 2, the calculation unit is, for example, the feature amount comparison unit 42 in FIG. 10, the sentence creation unit is, for example, the sentence configuration unit 43 in FIG. 10, the output control unit is, for example, Each can be configured by the input / output interface 13 of FIG.

In the image evaluation device of the present invention, the features of the input image are extracted, the adjectives and the features corresponding to the adjectives are retained, and the features of the extracted images and the features corresponding to the retained adjectives are retained. Is calculated, a sentence is created using the calculation result, and the created sentence is output.

According to the image evaluation apparatus of the present invention, the features of the input image are extracted, the adjectives and the features corresponding to the adjectives are held, and the features of the extracted images and the adjectives held are stored. The difference from the feature is calculated, a sentence is created using the calculation result, and the created sentence is output. Therefore, the evaluation of the input image is expressed as a sentient sentence using an adjective. be able to.

The text creating means of the image evaluation device of the present invention comprises:
An adjective having a feature close to the feature of the image extracted by the feature extracting unit can be selected from the calculation result of the calculating unit, and a sentence can be created using the selected adjective.

In the image evaluation apparatus of the present invention, an adjective having characteristics close to those of the image to be evaluated is selected, and a sentence is created using the adjective.

[0011] The image evaluation device of the present invention provides
The adjective input control means for controlling the input of adjectives is further provided, wherein the calculating means calls a characteristic of the adjective whose input is controlled by the adjective input control means from the holding means, and a difference from the characteristic of the image extracted by the characteristic extracting means. Can be calculated.

The adjective input means can be constituted by, for example, an input unit 51 shown in FIG.

In the image evaluation apparatus of the present invention, an adjective is input by a user, the characteristics of the input adjective are read, and compared with the characteristics of the image to be evaluated.

The text creating means of the image evaluation device of the present invention comprises:
Based on the calculation result of the calculation means, it is possible to create a sentence indicating a method for bringing the image to be evaluated closer to the feature of the adjective whose input is controlled by the adjective input control means.

In the image evaluation apparatus of the present invention, the difference between the image to be evaluated, the adjective, and the feature is calculated, and based on the result, a sentence indicating a method for bringing the image to be evaluated closer to the feature of the adjective. Is created.

The holding means of the image evaluation apparatus of the present invention classifies and holds the characteristics of the adjectives into a plurality of groups, and the calculating means calculates the characteristics of the adjectives of the desired group among the characteristics of the adjectives of the plurality of groups. To calculate the difference from the image.

In the image evaluation apparatus of the present invention, the characteristics of the adjectives are classified and held in a plurality of groups, and the difference from the image to be evaluated is calculated using the characteristics of the adjectives of the desired group.

[0018] The image evaluation apparatus of the present invention further comprises an image pickup means for picking up an image, and the feature extraction means can extract the features of the image taken by the image pickup means.

The imaging means is, for example, a camera 61 shown in FIG.
Can be configured.

In the image evaluation apparatus according to the present invention, an image is picked up and features of the picked-up image are extracted.

The output control means of the image evaluation apparatus of the present invention comprises:
The output of the text created by the text creation means via a network or a telephone line can be controlled.

In the image evaluation apparatus according to the present invention, the created text is output via a network or a telephone line.

[0023] The image evaluation apparatus of the present invention further comprises storage means for storing the contact information, and the output control means refers to the contact information stored in the storage means, and outputs the text created by the text creating means. Output can be controlled.

The storage means can be constituted by, for example, a contact database 65 shown in FIG.

In the image evaluation device of the present invention, the created text is output by referring to the contact information stored in the storage means.

[0026]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 shows a block diagram of an information processing apparatus to which the present invention is applied. The information processing device 1 is, for example, a personal computer.

In the information processing apparatus 1, a CPU (Centra
l Processing Unit) 11, the user can use a keyboard,
When a command is input using the input unit 12 including a mouse or the like, a signal representing the command is input via the input / output interface 13 and the internal bus 14. And C
The PU 11 reads the ROM (ReadOn) according to the input signal.
ly Memory) 15, HDD (Hard Disk Drive) 16, etc.
Adjective database input program described later using
An image evaluation program described later with reference to FIGS. 10 and 11, a design support program described later with reference to FIGS. 16 and 17, or a danger detection / notification program described later with reference to FIGS. Rand
om Access Memory) 17 for execution. RAM
17, in addition to the loaded program, for example,
Data and the like generated by the processing of the CPU 11 are stored.

The CPU 11 stores programs and various data input from other information processing devices via the network interface 18 and the internal bus 14 in the RAM 17. The information processing apparatus 1 can receive input of image data from a video camera or the like via the network interface 18, for example. Further, the CPU 11 transmits the processing result to, for example, the LC via the internal bus 14 and the input / output interface 13.
Monitor 19 composed of D (Liquid Crystal Display)
And output and display as necessary,
4. It can output to other information processing devices via the input / output interface 13 and the network interface 18.

FIG. 2 shows a functional block diagram when the adjective database input program is loaded into the RAM 17 and executed.

The image feature extracting unit 31 receives the input of the image, binarizes the input image, for example, as shown in FIG. 3A, and converts the ratio of the linear portion and the ratio of the curved portion. 3B, a color template is generated from the input image as shown in FIG. 3B, and the saturation and lightness of the image are obtained, or the input image is obtained as shown in FIG. 3C. By generating a color tone histogram from the image, the hue (whether cool or warm) of the input image, the number of colors used in the image, and the like are obtained.

The extraction results of the image features from the input image are as follows. For the hues, for example, warm colors when red, orange, and yellow are the most, cold colors, green, and green when the most are blue and blue-green.
When purple is the largest, the color is neutral, and the brightness is, for example, when white is 1.0 and black is 9.5, high brightness is 1.0 or more and less than 3.5, and 3.5 or more and 6. Medium brightness for less than 5,
A brightness of 6.5 or more and 9.5 or less is defined as low brightness.
For example, when high-saturation is evaluated on a 9-point scale, and low-saturation is evaluated on a 1-point scale, 1,2,3 are low saturation, less than 4,5,6 are medium saturation, 7,8,9 are high saturation. The degree of curve and the degree of linearity are, for example, large when the ratio of the curve or the ratio of the straight line in the line segment component obtained by binarization is 70% or more, and when the ratio is 30% or more and less than 70%. Medium or less than 30%, all feature amounts may be expressed in three stages, such as small, and numerical values of evaluation results may be used as they are for lightness, saturation, curve degree, linearity, and the like. Alternatively, each value may be represented using, for example, a membership function.

The correlation between an adjective and a feature is determined based on a questionnaire by a plurality of subjects. Therefore, the adjectives are classified and arranged for each category, and a questionnaire is conducted using a plurality of samples (for example, 50 sheets of each category, etc.) for each category or randomly regardless of the category. , Feature values (e.g., hue, brightness, saturation, number of colors,
The degree of curve and the degree of linearity are calculated, and the relationship between each adjective and the feature is quantified. The adjective-image feature quantity correlation calculation unit 32 receives the result of the questionnaire for each image and obtains a correlation with the feature quantity of the image input from the image feature quantity extraction unit 31.

FIG. 4 shows a sample used for the questionnaire and an example of the result of the questionnaire. Here, an example is shown in which a questionnaire regarding the “cuteness” of each sample is given to the adjective “cute”.

Here, the method of evaluating “pretty” for each sample is, for example, asking the subject to answer whether or not the sample is “cute” and quantifying the proportion of the subject who answered “cute”. Good, "cuteness"
May be evaluated by a subject on a five-point scale, and the results may be averaged and converted into a numerical value, or may be evaluated using other methods.

The result of the questionnaire and the image feature quantity extraction unit 3
FIG. 5 shows an example of an extraction result of the feature amount of the sample image data used in the questionnaire extracted in 1. Here, the feature amount is evaluated in three stages, such as large, medium, small, and high, medium, low, respectively. However, for example, the feature amount may be quantified using a membership function or the like (in this case, A method such as fuzzy inference can be used for the comparison calculation between the feature amount of the adjective and the feature amount of the image to be evaluated.)

FIG. 6 shows an example of a correlation graph for obtaining a correlation between an adjective and each feature amount, which is generated using the questionnaire result of the sample image and the extraction result of each feature amount. For example, in FIG. 6A, the correlation between the adjective "cute" and the brightness is shown, and it can be seen that the higher the brightness, the more subjects who feel "cute". Similarly, in FIG. 6A, the correlation between the adjective "cute" and the degree of curve is shown, and it can be seen that the higher the degree of curve, the more subjects who feel "cute".

The rule generator 33 generates an adjective based on the correlation calculation result of the adjective-image feature quantity correlation calculator 32 (for example, the correlation between the adjective and the feature derived from the correlation graph described with reference to FIG. 6). Then, a state in which each feature amount shows the highest correlation is calculated, and a rule for each feature amount corresponding to each adjective as shown in FIG. 7 is generated, for example.

In the questionnaire result in this case,
Images that feel "cute" have warm hues,
An image with high lightness, low saturation, medium number of colors, large curveness and low linearity, and an image that feels "lonely" has a cool hue, low brightness, and low saturation. Is low, the number of colors is low, and the degree of curve and linearity is moderate.
It can be seen that the hue is warm, the brightness is high, the saturation is high, the number of colors is large, the curve is small, and the linearity is large.
An image that feels "soft" has a warm hue,
It can be seen that the brightness is medium, the saturation is medium, the number of colors is small, the curve is large, and the linearity is small. For example, if there is no correlation between a certain feature and an adjective, the corresponding column may be left blank.

Here, the rule corresponding to each adjective is generated in three stages. However, the rule may be generated in smaller stages, may be digitized, or a membership function may be used.

In the adjective database 34, an adjective list classified by category in which adjectives are classified by category is registered, and a corresponding adjective list in the adjective list classified by category is
Rules of each feature amount corresponding to each adjective generated by the rule generation unit 33 are registered.

FIG. 8 shows an example of a category-specific adjective list registered in the adjective database 34. For example, “Cold and warm” and “Atmosphere” are registered in the middle category for the large category of “Nature”, and the adjectives corresponding to each are “Cold”, “Warm” and “Lonely”, “Lively”, When “quiet”, “loud”, “bright”, and the like are registered in advance, the rules of the feature amounts for each adjective input from the rule generation unit 33 are registered in the corresponding columns. If the major classification, the middle classification, and the minor classification are not registered in advance, the major classification and the middle classification are left blank, and a category-specific adjective list in which each adjective is classified into the minor classification is generated. The rule of the feature amount for each adjective may be registered.

Next, referring to the flowchart of FIG.
The input processing of the adjective database will be described.

In step S1, the image feature quantity extraction unit 31 determines whether or not the sample image used for the questionnaire has been input. If it is determined in step S1 that a sample image has not been input, the process of step S1 is repeated until it is determined that a sample image has been input.

If it is determined in step S 1 that a sample image has been input, in step S 2, the image feature quantity extraction unit 31 performs, as described with reference to FIG.
For example, by binarizing or creating a color tone histogram, the feature amount (for example,
The hue, lightness, saturation, number of colors, curveness, and linearity described with reference to FIG. 5 are extracted and output to the adjective-image feature quantity correlation calculation unit 32.

In step S3, the adjective-image feature quantity correlation calculator 32 determines whether a questionnaire result for the sample image has been input. In step S3, when it is determined that the questionnaire result has not been input, until it is determined that the questionnaire result has been input,
Step S3 is repeated.

If it is determined in step S3 that the questionnaire result has been input, then in step S4,
The adjective-image feature value correlation calculation unit 32 evaluates the characteristics of each sample image input from the image feature value extraction unit 31 in step S2 and the “adjective likeness” of the sample image obtained by the questionnaire (for example, cuteness). For example, the correlation with the value evaluation is calculated by creating the correlation graph described with reference to FIG.

In step S5, the rule generation unit 33
Generates, for example, a rule of the feature amount for each adjective as described with reference to FIG. 7 based on the correlation between the feature amount and the adjective input in step S4, and outputs the rule to the adjective database.

In step S6, the adjective database 34 determines whether a category-specific adjective list as described with reference to FIG. 8, for example, is registered.
If it is determined in step S6 that the category-specific adjective list is registered, the process proceeds to step S8.

If it is determined in step S6 that the category-specific adjective list is not registered, in step S7, the adjective database 34
Based on the rule of the feature amount for each adjective input in 5, a category-specific adjective list in which the large classification and the middle classification are blank (that is, only the small classification) is generated.

In step S8, the adjective database 34 registers the rule of the feature amount for each adjective in the category-based adjective list, and the process ends.

The adjectives registered in the adjective database 34 and the feature amounts corresponding to the adjectives are registered by the method described with reference to FIGS. May be registered by using another method such as manually inputting the password. Also,
A plurality of category-specific adjective lists can be registered in the adjective database 34. In this case, each category-specific adjective list can be identified by attaching a group name or the like to each category-specific adjective list.

FIG. 10 is a functional block diagram when the image evaluation program is loaded into the RAM 17 and executed.

The image feature amount extraction unit 41 extracts the feature amount of the input image and outputs the same to the feature amount comparison unit 42, similarly to the image feature amount extraction unit 31 described with reference to FIG. In the adjective database 34, correlation rules between feature amounts and adjectives are registered in a category-based adjective list by the adjective database input processing described with reference to FIG.

The feature amount comparing section 42 is provided with the image feature amount extracting section 4.
Calculating a difference between the feature of the selected adjective and the feature of the input image based on the feature of the image input from 1;
An adjective having similar characteristics is selected from the adjectives registered in the adjective database 34. Here, when the adjectives registered in the adjective database 34 are classified by the large classification and the middle classification of the category list, for example, one adjective having the closest feature is selected for each middle classification. You may do it. Then, the feature value comparison unit 42 calculates “adjectiveness” as “weight” based on the difference between the feature value of the selected adjective and the feature value of the input image, and outputs the weight to the sentence composition unit 43. The sentence construction unit 43 creates and outputs a sentence in accordance with the adjective weight input from the feature amount comparison unit 42 (the details of the method of calculating the adjective weight and the method of creating the sentence will be described later).

Next, the processing of the image evaluation processing program will be described with reference to the flowchart of FIG.

In step S11, the image feature quantity extraction unit 41 determines whether an image to be evaluated has been input. If it is determined in step S11 that an image has not been input, the process of step S11 is repeated until it is determined that an image has been input.

If it is determined in step S11 that an image has been input, in step S12, the image feature amount extraction unit 41 performs, for example, binarization or color tone histogram generation as described with reference to FIG. By doing so, the feature amounts of the input image (for example, hue, lightness, saturation, number of colors, curve degree, and linearity described with reference to FIG. 5) are extracted and output to the feature amount comparison unit 42. .

In step S13, the feature amount comparison unit 4
Reference is made to the category-specific adjective list generated by the processing described with reference to FIG. 9 and registered in the adjective database 34, and the difference between each feature amount of the input image and the registered adjective is calculated. Calculate and, based on the calculation result,
An adjective representing the evaluation of the image (that is, an adjective having a feature close to the feature of the input image) is selected, and “weight” representing the adjective likelihood is calculated and output to the text forming unit 43. An example of selecting an adjective and calculating a weight will be described with reference to FIG.

For example, as shown in FIG. 12, consider a coordinate space using feature amounts (for example, curve degree and saturation) as coordinate axes. In the coordinate space, a predetermined position of the image to be evaluated is determined based on each feature amount extracted in step S12. Then, from the position in the coordinate space of the image to be evaluated, a certain adjective (for example, “cute”
By calculating the distance to the position in the coordinate space indicated by the adjective (), it is possible to quantify the degree to which the image to be evaluated is represented by the target adjective (for example, the degree of “cuteness”). Becomes

For example, the image P to be evaluated has a saturation of 0.
5. If the degree of curve is 0.5, the adjective database 3
The adjective "cute" registered in No. 4 (saturation 0.8
5, curve degree 0.35) and "beautiful" (saturation 0.3
5, the distance from the curve degree 0.85) is 0.38. As an adjective representing the evaluation of the image P, an adjective having a short distance in the coordinate space is selected. Further, the closer the distance is, the more similar the feature of the adjective is, that is, the larger the weight is. In FIG. 12, the positional relationship between the image to be evaluated and the adjective is described using a two-dimensional space having “curvature” and “saturation” as coordinate axes. Is calculated in the multidimensional coordinate space.

In step S14, the sentence construction unit 43
Uses the adjective corresponding to the evaluation target image and its weight, which are input from the feature amount comparison unit 42 in step S13, to form a sentence indicating the kansei evaluation for the evaluation target image.

For each adjective, a modifier is predetermined for each weight, as shown in FIG. This is because, for each adjective, a modifier that can be used to represent the degree of the likelihood of the "adjective" is different, for example, "cold"
In the case of, the degree of cold is "slightly cold",
Qualified as "slightly cold", "cold", "substantially cold", and "lonely" means "slightly lonely", "slightly lonely", "lonely", "very sad" , And "quiet" means "slightly quiet", "quiet", "really quiet"
Qualified as "very quiet".

The composed sentence is, for example, “(middle classification)
Is (modifier) + (adjective). "When two or more adjectives are applicable to one intermediate classification, for example, when two adjectives are applicable,
It is said that "(the middle class) feels (modifier) + (adjective) or (modifier) + (adjective)".

For example, the calculation results of the feature weights of the images whose major classification is “natural” and whose middle classifications are “cold” and “atmosphere” are 1.0 for “cold” and 0 for “lonely”.
5. If 0.8 for "quiet", the sentence says, "I feel quite cold when it comes to cold and warm. I feel a little lonely and really quiet about the atmosphere." Become.

In step S15, the sentence forming unit 43
Outputs the text composed in step S14, and the process ends. That is, the CPU 1 of the information processing apparatus 1
1 is an internal bus 14 and an input / output interface 13
And outputs the composed sentence to the monitor 19 via. For example, as shown in FIG. 14, an image to be evaluated may be displayed together with the composed text so that the user can easily understand the evaluation target and the evaluation result.

As described above, the evaluation of an image can be output not as a numerical evaluation for each feature amount but as a sentence indicating an emotional evaluation using adjectives.

Here, for example, when designing a poster or the like, it is possible to design a poster that seems to be “cute” or to design a poster that is “buzzy” and feels “bright”. Therefore,
An adjective representing the target image's kansei evaluation is input, and the feature amount comparison unit 42 calculates the distance between the position indicated by the feature extracted from the designed image and the position of the target adjective in the feature amount coordinate space. (That is, the difference from the target adjective) is calculated, and from the calculation result, a sentence that advises how to correct the text is created and output, thereby supporting design. . In addition, for example, by preparing a category-specific adjective list in which the questionnaire results are classified by gender, age, and occupation in the adjective database 34, for example, it is possible to evaluate posters and the like for each corresponding target. Become.

FIG. 15 shows that the design support program is RA
FIG. 4 shows a functional block diagram when loaded into M17 and executed. Parts corresponding to those in FIG. 10 are denoted by the same reference numerals, and description thereof will be omitted as appropriate. In this case, the configuration is basically the same as that in FIG. 10 except that an input unit 51 is newly provided and an advice creation unit 52 is provided instead of the text composition unit 43.

The adjective database 34 has prepared adjective lists by category for each target (for example, gender, age, and occupation). Each adjective list by category can be identified by a group name corresponding to the target. is there. The target adjective and the target group name input by the user using the input unit 12 of FIG. 1 are input to the feature amount comparison unit 42 via the input unit 51. The feature amount comparing unit 42 reads out the feature amount corresponding to the target and the adjective inputted from the input unit 51 from the adjective database 34, and outputs the feature of the image to be evaluated inputted from the image feature amount extracting unit 41 and the target The difference from the adjective to be performed is obtained and input to the advice creation unit 52.
The advice creating unit 52 brings the evaluation of the image to be evaluated closer to the target adjective based on the difference between the feature of the image to be evaluated input from the feature amount comparing unit 42 and the target adjective. Create and output advice.

Next, the design support processing will be described with reference to the flowchart of FIG.

In steps S21 and S22, the same processing as in steps S11 and S12 in FIG. 11 is performed.
The features of the input image to be evaluated are extracted and output to the feature amount comparing unit 42.

In step S23, the feature amount comparing unit 4
2 determines whether or not a target adjective and a target have been input from the input unit 51. If it is determined in step S23 that the target adjective and target have not been input, the process proceeds to step S2 until it is determined that the target adjective and target have been input.
Step 3 is repeated.

If it is determined in step S23 that the target adjective and target have been input from the input unit 51, then in step S24, the feature quantity comparing unit 4
2 reads from the adjective database 34 the target adjective feature quantity of the category-specific adjective list in which the questionnaire result of the desired target is registered, which is input from the input unit 51, and the evaluation input in step S22. Compare with the features of the target image. That is, the feature amount comparison unit 42 calculates the distance between the image to be evaluated in the multidimensional coordinate space and the target adjective, using the respective features of the image and the adjective as coordinate axes, described with reference to FIG. Find the difference for each quantity. Feature amount comparison unit 4
2 outputs to the advice creation unit 52 the distance between the evaluation target image in the coordinate space using the feature amount as the coordinate axis and the target adjective, and the difference for each feature.

In step S25, the advice creating section 52 evaluates the image and approaches the target adjective based on the distance between the image to be evaluated and the target adjective input in step S24 and the difference for each feature. Create advice for.

The sentence representing the evaluation of the image is, for example, “(target) is (modifier) + (adjective) + (“ feels ”or“ does not feel ”)”. For example, if the distance from the evaluation target image targeting "housewife" and targeting the adjective "cute" is 0.8 or more,
The sentence "The housewife does not feel cute" is created, and if it is 0.4 or more and less than 0.8, the sentence "The housewife does not feel cute" is created and 0.1 or more. If the number is less than 4, a sentence "Housewife feels cute" is created. If the number is less than 0.1, a sentence "Housewife feels very cute" is created.

The advice for bringing the evaluation target image closer to the target is, for example, based on the difference for each feature, for example,
"(Feature) (instruction) + (feature) (instruction) + ... will make you feel like (adjective)." For example, if the difference between the lightness and the curve degree is large, an advice is created that “It is possible to feel cute by increasing the lightness and increasing the curve degree”.

In step S26, the advice creating section 52 outputs the evaluation and advice sentences created in step S25, and the process ends. That is, the CPU 11 of the information processing apparatus 1 outputs the created text to the monitor 19 via the internal bus 14 and the input / output interface 13. Further, an image to be evaluated may be displayed on the monitor 19 at the same time as the created text.

Here, the case where one target and one target adjective are input and evaluated has been described. However, a plurality of target and target adjectives may be input and each may be evaluated.

Further, for example, an image of the platform of the station, etc.
Take an image using a video camera or the like, evaluate the image,
If it is determined to be dangerous, it can be reported to a remote location. FIG. 17 shows a functional block diagram when the danger detection / notification application is loaded on the RAM 17 and executed.

Parts corresponding to those in FIG. 10 are denoted by the same reference numerals, and description thereof will be omitted as appropriate. In this case, the camera 61 and the judgment unit 6 are newly added.
2. A report control unit 64 and a contact database 65 are provided. The report control unit 64 includes a LAN 66 and a telephone line 67.
The configuration is basically the same as that in FIG.

The camera 61 is installed, for example, on the platform of a station, captures an image of the platform, and outputs the captured image to the image feature quantity extraction unit 41. The determining unit 62 determines whether or not there is danger based on the sentence input from the sentence composition unit 43, and when it is determined to be dangerous, causes the notification control unit 64 to refer to the contact information from the contact information database 65, An external notification is made via the LAN 66 or the telephone line 67.

Next, the danger detection / notification process will be described with reference to the flowchart of FIG.

In step S 31, the image feature extracting unit 41 extracts the feature of the image input from the camera 61 and outputs the feature to the feature comparing unit 42.

In step S32, the feature amount comparing section 4
2 refers to the data registered in the adjective database 34, executes the same processing as in step S13 in FIG. 11, compares the feature amount of the input image with the feature amount of the adjective, and compares the comparison result with a sentence. Output to the configuration unit 43. Here, in the adjective database 34, adjectives indicating congestion, such as "crowded", "vacant", "busy", and the like, and feature amounts corresponding to the adjectives are registered.
Based on the input comparison result, the text composition unit 43
By executing the same processing as in step S14, a sentence is formed and output to the determination unit 62.

In step S33, the judgment section 62
From the input text, the degree of congestion of the home is determined, and whether danger is detected or not is determined. That is, if the input sentence is "very empty" or "empty",
If the home is not crowded, and if it is "busy", "very busy", or "very busy", the home is judged to be busy and dangerous. If it is determined in step S33 that danger has not been detected, the process proceeds to step S35.

If it is determined in step S33 that danger has been detected, in step S34, the determination unit 62 outputs a signal indicating that danger has been detected to the notification control unit 64, and the notification control unit 64 , And refers to the contact database 65 to search for a corresponding contact (for example, a station manager's office) and reports via the LAN 66 or the telephone line 67.

In step S35, the judgment section 62
The result of the determination is output, and the process ends. That is, the CPU 11 of the information processing device 1 outputs the composed text to the monitor 19 via the internal bus 14 and the input / output interface 13. Further, an image captured by the camera 61 may be displayed on the monitor 19 at the same time as the composed text.

As described above, it is possible to evaluate “congestion degree” using the comparison between the image and the adjective, and report the evaluation result to a remote place.

[0090]

According to the present invention, the evaluation of an input image can be expressed as an emotional sentence using adjectives.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of an information processing apparatus to which the present invention has been applied.

FIG. 2 is a functional block diagram in the case where adjective database input software is loaded into the RAM of FIG. 1 and executed.

FIG. 3 is a diagram illustrating an example of a sample image and a questionnaire result for the adjective “cute”.

FIG. 4 is a diagram for explaining processing of an image feature amount extraction unit in FIG. 2;

FIG. 5 is a diagram illustrating an input sample image, a questionnaire result, and a feature amount extraction result.

FIG. 6 is a diagram for explaining a correlation between an adjective and a feature amount of an image.

FIG. 7 is a diagram illustrating an example of a rule for each adjective, generated by the rule generation unit in FIG. 2;

FIG. 8 is a diagram showing an example of a category-specific adjective list registered in an adjective database.

FIG. 9 is a flowchart for explaining adjective database processing.

FIG. 10 is a functional block diagram in a case where image evaluation software is loaded into the RAM of FIG. 1 and executed.

FIG. 11 is a flowchart illustrating an image evaluation process.

FIG. 12 is a diagram for explaining a method of determining a weight between a target image and an adjective.

FIG. 13 is a diagram for describing modifiers for each adjective when constructing a sentence.

FIG. 14 is a diagram illustrating an example of display of an evaluation result.

FIG. 15 is a functional block diagram when the design support software is loaded on the RAM of FIG. 1 and executed.

FIG. 16 is a flowchart illustrating a design support process.

17 is a functional block diagram when danger detection / notification software is loaded into the RAM in FIG. 1 and executed.

FIG. 18 is a flowchart illustrating a danger detection / notification process.

[Explanation of symbols]

 Reference Signs List 1 information processing device 11 CPU 12 input unit 13 input / output interface 19 monitor 20 network interface 31 image feature amount extraction unit 32 adjective-image feature amount correlation calculation unit 33 rule generation unit 34 adjective database 41 image feature amount extraction unit 42 feature amount comparison Part 43 Text composition part 52 Advice creation part 61 Camera 62 Judgment part 64 Report control part 65 Contact database

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5B057 DA20 DB02 DB06 DC09 DC25 5L096 AA02 EA39 FA34 FA68 FA81 GA41

Claims (8)

[Claims] 1. An image evaluation apparatus for evaluating an input image, a feature extracting unit for extracting a feature of the input image, a holding unit for holding an adjective and a feature corresponding to the adjective, and the feature extraction. Features of the image extracted by the means,
Calculating means for calculating a difference from a feature corresponding to the adjective held in the holding means; a text creating means for creating a sentence by using a calculation result by the calculating means; and a sentence created by the sentence creating means And an output control means for controlling the output of the sentence. 2. The sentence creation means selects an adjective having a feature close to the feature of the image extracted by the feature extraction means from a calculation result of the calculation means, and uses the selected adjective to select a sentence. The image evaluation device according to claim 1, wherein 3. An adjective input control means for controlling an input of the adjective by a user, wherein the calculating means calls a characteristic of the adjective whose input is controlled by the adjective input control from the holding means, The image evaluation device according to claim 1, wherein a difference from a feature of the image extracted by the feature extraction unit is calculated. 4. The sentence creating means, based on a calculation result of the calculating means, generates a sentence indicating a method for bringing the image closer to a feature of the adjective whose input is controlled by the adjective input control means. The image evaluation device according to claim 3, wherein the image evaluation device is created. 5. The holding means classifies and holds the characteristics of the adjectives in a plurality of groups, and the calculating means calculates the characteristics of the adjectives in a desired group among the characteristics of the adjectives in the plurality of groups. The image evaluation device according to claim 1, wherein a difference from the image is calculated by using the image evaluation device. 6. The image processing apparatus according to claim 1, further comprising an imaging unit configured to capture an image, wherein the feature extraction unit extracts a feature of the image captured by the imaging unit. Image evaluation device. 7. The output control means controls output of the text created by the text creation means via a network or a telephone line.
7. The image evaluation device according to any one of claims 1 to 6. 8. A storage unit for storing contact information, wherein the output control unit refers to the contact information recorded in the storage unit, and outputs an output of the text created by the text creation unit. 2. The method according to claim 1, wherein the control is performed.
An image evaluation device according to any one of claims 1 to 7.