JP2001236508A - Device and method for calculating remarkability - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device and a method for calculating remarkableness, which can calculate the remarkableness of an image component while considering the degree of human experience or memory when calculating remarkableness. SOLUTION: This is a remarkableness calculating method for inputting first image data to become the object of remarkableness calculation, extracting the features of respective parts in the first image data, quantifying the remarkableness on the basis of the extracted features of respective parts and displaying outputting the calculated remarkableness, a plurality of second image data to become learning objects are inputted, the kinds of features are previously learnt and determined on the basis of a plurality of second image data and the features of respective parts are extracted on the basis of the determined kinds of features and the first image data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus and a method for calculating the saliency of each part of an image as a scale for measuring the importance of each part in a displayed image.

[0002]

2. Description of the Related Art When observing an image, a person often pays attention to a specific portion in the image without being particularly conscious. This is often the case when the part has different characteristics than the other parts. In this way, an evaluation scale indicating the degree to which a person's eyes are more likely to pay attention when observing an image is referred to as “prominence”.

For example, when observing an image as shown in FIG. 1, attention of a human eye is directed to a line inclined only by one, so that this part is more prominent than other parts. Is considered high.

Various methods have been considered as methods for quantifying and calculating such saliency. See, for example, "A model of Silency-based Visual Atten," by L. Itti et al.
tion for Rapid Scene Analysis ", IEEE Transactions o
n Pattern Analysis and Machine Intelligence 1998; 2
0 (11): 1254-1259.)), The local variation of brightness, color, line direction, etc. of each part of the image is calculated as a feature of each part, and the weighted average value is calculated to obtain the saliency. Is defined. For example, in the case of FIG. 1, the amount of variation in the line direction is largely evaluated at a portion of a line inclined only by one line. A similar approach was taken by Wolff (JMWolf)
e) "A modified model of visual search (" Guided sea
rch 2.0: arevised model of visual search ", Psychonom
ic Bulletin Review, vol.1, pp.202-238, 1994.) ”.

FIG. 2 is a flow chart of a process in such a conventional saliency calculation method. First, an image as a saliency calculation target is input (step S21), and a variation in a line direction (a difference from a surrounding line direction) or the like is calculated as a feature of each unit in each unit of the image (step S22). Then, saliency is calculated based on those features (step S23),
Output and display using a saliency map or the like (step S2)
4). Here, the “saliency map” refers to a general term for the saliency of a quantified image displayed on the same coordinates as the image for which the saliency is calculated using the shades of color.

[0006]

However, in the above-described method of calculating saliency, the type of feature to be extracted from an image is fixed, and the saliency is most accurately determined by evaluating the saliency with any feature. There is a problem that it does not matter whether it can be judged.

That is, since saliency is calculated based on one or more types of features specified in advance, even if an image from which saliency is to be extracted has a part clearly different from the surroundings, This is because if it is different from the type of the specified feature, a problem occurs that it is not determined that the feature is high.

In addition, the above-described calculation method has a problem that saliency cannot be calculated in consideration of factors such as human experience and memory.

For example, in FIG. 3 (a), it can be easily found that there is only one "2" in the character "5" which is commonly used. That is, it is easy to determine that the character “2” has high saliency.

On the other hand, FIG. 3 (b) finds the same character "2" as in FIG. 3 (a), although it is an image obtained by rotating the same image as in FIG. 3 (a) by 90 degrees. This is more difficult than in the case of FIG. That is, by rotating the image itself by 90 degrees, '5'
It is considered that the reason is that characters such as and '2' have become figures that are not usually seen, and it is considered that human experience and memory greatly influence the judgment of saliency.

In FIG. 4A, it is easy to find an inverted character of "N" from "N" which is a character which is commonly used. On the other hand, in FIG. 4B in which the entirety of FIG. 4A is inverted left and right, finding a character “N” that is usually familiar from the inverted characters of “N” that is not usually seen is not performed in FIG. This is more difficult than the case of a). Therefore, also in this example, it is considered that human experience and memory greatly influence the determination of saliency.

The present invention provides an apparatus and method for calculating saliency, which can be calculated in consideration of human experience and the degree of memory when calculating the saliency of an image component in order to solve the above problems. The purpose is to do.

[0013]

In order to achieve the above object, a saliency calculating apparatus according to the present invention comprises: an image input unit for inputting first image data to be calculated;
A feature extraction unit that extracts features of each part of the image data, a saliency calculation unit that quantifies saliency based on the extracted features of each part, and an output display unit that displays and outputs the calculated saliency. A learning image group input unit for inputting a plurality of second image data to be learned, and learning and determining a feature type in advance based on the plurality of second image data. And a feature type determining unit that inputs the type of the feature determined by the feature type determining unit to the feature extracting unit. The feature extracting unit determines a feature of each unit based on the feature type and the first image data. It is characterized by extracting.

With this configuration, the type of feature to be extracted from the learning image can be changed, and the saliency can be determined based on the type of feature that matches the image data. Therefore, by emulating such a learning effect with the degree of human experience and memory, it is possible to determine the saliency more similar to the human sense.

Further, the saliency calculating device according to the present invention comprises:
The image processing apparatus further includes an area extracting unit that extracts third image data, which is an area with high saliency, from the first image data based on the saliency output from the output display unit. Further includes an extraction region re-input unit for inputting the third image data extracted by the region extraction unit, wherein the feature type determination unit learns the type of the feature based on the third image data, and the feature extraction unit extracts Preferably, the type of feature to be re-determined is determined. This is because a new type of feature can be set based on an image for which saliency has been determined, and thus, by repeatedly calculating saliency, it is possible to determine saliency similar to a human sense.

Further, the saliency calculating device according to the present invention comprises:
It is preferable to use KL conversion in the feature type determination unit. Furthermore, in the saliency calculation device according to the present invention, it is preferable that the saliency calculation unit uses KL conversion. This is because this is one method for more faithfully expressing image data using a limited number of features.

Further, the present invention is characterized by software for executing the function of the saliency calculating device as described above as a processing step of a computer.
A step of inputting first image data to be subjected to saliency calculation, a step of extracting characteristics of each part of the first image data, and a step of quantifying saliency based on the extracted characteristics of each part. Displaying the calculated saliency and outputting the calculated saliency, wherein a plurality of second image data to be learned are input, and a feature is calculated based on the plurality of second image data. A method of preliminarily learning and determining the type, a method of extracting the characteristic of each part based on the determined type of the characteristic and the first image data, and a computer recording such a step as a program It is a readable recording medium.

With this configuration, by loading and executing the program on the computer, the type of feature to be extracted by the learning image can be changed, and the saliency can be increased based on the type of feature matching the image data. You can judge. Therefore, it is possible to realize a saliency calculating device that can judge saliency more similar to a human sense by simulating the learning effect with the degree of human experience and memory.

[0019]

(Embodiment 1) Hereinafter, a saliency calculating apparatus according to Embodiment 1 of the present invention will be described with reference to the drawings. FIG. 5 is a configuration diagram of the saliency calculating device according to the first embodiment of the present invention. In FIG.
Reference numeral 51 denotes an image input unit, 52 denotes a feature extraction unit, 53 denotes a saliency calculation unit, and 54 denotes an output display unit. As far as this configuration is concerned, there is no difference from the conventional saliency calculation device. do not do.

In the first embodiment, at the time of feature extraction,
This is different from the conventional example in that features are determined based on the types of features learned in advance from a plurality of image groups, that is, the types of features to be extracted vary depending on the images to be learned. That is, the feature is that a learning image group input unit 521 and a feature type determination unit 522 are provided to realize such a function.

The saliency calculating device according to the first embodiment first inputs an image to be extracted for saliency from an image input unit 51 in FIG. The input medium is generally a scanner, a digital camera, or the like, but is not limited thereto. Further, a bitmap image, a TIFF file, a JPEG file, or the like stored in a file may be used.

Next, a feature is extracted from the saliency calculation target image by the feature extracting unit 52. In the first embodiment, the feature type determining unit 522 determines the type of the feature to be extracted in advance. You can decide.

That is, first, an image to be learned is input from the learning image group input unit 521. For example, a plurality of images, such as an image in which an alphabet is described, an image in which hiragana or katakana is described, are input using an input medium such as a scanner in the same manner as the image input unit 51. The type of the feature determined to have the feature in the image input from the learning image group input unit 521 is:
This corresponds to experience, memory, and the like when a person judges the saliency of an image.

The feature type determining section 522 determines the type of feature for which saliency should be calculated based on the input image group. Specifically, the following processing is performed.

First, a local region having a characteristic is detected from the entire input learning image by a predetermined method. For this purpose, for example, a window region having a partial size of the learning image is determined, and a local region having a feature is detected from the entire learning image while moving the window region from the upper, lower, left, and right ends of the learning image in dot units or the like. However, the method is not particularly limited to this method.

The image pattern in each local area is regarded as a vector, and a covariance matrix is calculated based on all the vectors corresponding to each local area. Here, the covariance matrix P refers to a matrix that can be expressed by (Equation 1), and n
From m-dimensional vector v _i, is intended to be calculated to determine the degree of dispersion.

[0027]

(Equation 1)

In equation (1), the vector v _ave is n
The simple average of m-dimensional vector v _i, subscript 'T' is that it is a transposed matrix, respectively. Therefore, the obtained covariance matrix P is a square matrix with m rows and m columns.

The type of feature to be selected is obtained by selecting, from the eigenvectors of the covariance matrix, the vector having the largest eigenvalue until the sum of the eigenvalues reaches a certain threshold value. The eigenvectors e ₁ , e ₂ ,..., E _q thus selected are referred to as main eigenvectors.

Therefore, the image pattern of each local region of the learning image is approximated by a weighted linear sum of these q main eigenvectors. That is, in order to obtain a weighted linear sum, a main eigenvector of each image pattern is obtained as a weight by KL expansion, and an inner product Q of a vector v representing the image pattern and each main eigenvector e _i (i is a natural number from 1 to q) is calculated. By calculating as in (Equation 2), q feature vectors are determined.

[0031]

(Equation 2)

In the conventional method, the main eigenvector is fixed, and the type of feature serving as a reference for calculating saliency is fixed. On the other hand, according to the first embodiment, the main eigenvector depends on the type of sample image. fluctuate. That is, if the sample image is an image displaying the alphabet, the alphabet is selected as one of the feature vectors.

The KL expansion is only one method for expressing an image pattern as faithfully as possible with a limited number of features, and is a method for approximately expressing a large number of learning patterns with a limited number of features. If so, any method may be used.

The q inner products thus calculated are regarded as types of features in each local region, and as a q-dimensional feature vector, a pattern similar to the image pattern of a certain local region of the input image is assumed to be a pattern of many learning patterns. If it appears frequently in the image, this image pattern can be reproduced relatively faithfully by the extracted features.

The fidelity of reproduction can be evaluated by the difference between the square norm of the image pattern and the square norm of the feature vector. The difference represents a residual when the image pattern is represented by a feature vector, and the smaller the value, the more faithfully the image can be reproduced.

Next, the saliency calculating section 53 calculates the saliency based on the feature selected based on the obtained main eigenvector. Specifically, the following processing is performed.

[0037] First, the KL transform all feature vectors extracted from the local region as a target, likewise the r to that described above leading eigenvectors e _1, e _2, ..., seek e _r. Next, an inner product of the feature vector extracted from each local region and each of the r main eigenvectors is calculated as a secondary feature vector.

The difference between the square norm of the r secondary feature vectors obtained in this way and the square norm when the image pattern of each local region is regarded as a vector is calculated to indicate the saliency of each local region. Use as an index.

For example, as shown in FIG. 6, if the vector 61 of the image pattern is n-dimensional (in FIG. 6, it is represented in a three-dimensional space for simplification), the feature vector 6
Finding 2 corresponds to projecting onto a q-dimensional space (in FIG. 6, a two-dimensional plane is used for simplification). Then, obtaining the secondary feature vector 63 corresponds to projecting a point in the q-dimensional space onto an r-dimensional space (in FIG. 6, a straight line is used to simplify the description). That is, each time the KL conversion is performed, the space is converted into a space having a small dimension. Then, the difference 64 of the square norm to be obtained is a thick line portion indicating the difference between the vector 61 of the image pattern and the secondary feature vector 63.

The saliency obtained by such processing represents a residual when an image pattern of a certain local region is approximated by a weighted linear sum of q main eigenvectors determined from a feature set of patterns in all local regions in the input image. ing. That is, when the image pattern of a certain local area is not similar to the pattern of the other area, the saliency becomes high. Also, the saliency increases when the image pattern of a certain local area is not similar to the local pattern in the learning image.

As a pre-process for extracting features from an image, a simultaneous occurrence matrix O may be calculated in each local region. The co-occurrence matrix O refers to a matrix represented by (Equation 3), and has a pixel value I (x,
y) and the pixel value I (x
+ I, y + j).

[0042]

(Equation 3)

By performing the pre-processing using such a matrix, it is possible to minimize fluctuations in image pattern recognition due to parallel movement. For example, as shown in FIG. 7, when the image is large and part of the image is recognized by the window, the character “5” is recognized to be located on the right side of the window in the window (a), whereas In (b), the character “5” is recognized as being located on the left side of the window. Recognition of these as different images is a simple translation, and even images that should be recognized as the same image pattern may be recognized as having different features. In the limited first embodiment, many features are extracted, which is not preferable.

Therefore, by calculating the co-occurrence matrix O in such a manner that the pattern is not recognized as another pattern depending on the change in the window position due to the operation of the window position and the value does not change even if the window is moved in parallel, a few features are obtained. , It is possible to extract the essential difference.

Finally, the output display unit 54 outputs and displays the saliency calculated as the saliency map. For example, FIG. 8 and FIG. 9 are illustrations of an output result of actually calculating saliency after obtaining a co-occurrence matrix. 8 and 9 show output displays corresponding to FIGS. 3 and 4, respectively.

First, in FIG. 8A, characters with high saliency are displayed in white. In FIG. 8A, since there is only one place with high saliency,
By paying attention to that part, the character '2' can be easily found.

On the other hand, in FIG. 8B, since the saliency of all the characters is almost the same, attention is drawn to each character in turn, and the character of "2" is It turns out that it takes time to find out.

Similarly, in FIG. 9A, characters with high saliency are displayed in white. On the other hand, in FIG. 9B, it is apparent that it is difficult to find the character 'N' since the saliency is high in the inverted character portion of 'N' which is not the character 'N'. is there.

The processing flow of the program for realizing the saliency calculating device according to the first embodiment of the present invention is shown in FIG.
0.

First, before inputting an image to be subjected to saliency calculation, a learning image group to be sampled is input (step 101), and a covariance matrix is calculated based on all vectors corresponding to each local region of each image. (Step S10)
2). Then, the main eigenvector is determined as a feature type by performing KL transformation or the like based on the covariance matrix (step S103).

After the type of the feature is determined, an image to be subjected to saliency calculation is input (step S104). Then, first, a feature vector is calculated based on the main eigenvector (step S105). Next, a main eigenvector is determined again by KL conversion or the like based on the calculated feature vector (step S106), and a secondary feature vector is calculated (step S107).

Finally, the difference of the square norm between the vector corresponding to each local region of each image and the secondary feature vector is calculated (step S108), and output and displayed as a saliency map (step S109).

On the other hand, the above-described method is a method of preparing a large number of learning image groups in advance.
It is also conceivable to extract a partial region of an image based on the calculated saliency and gradually update the feature type based on the extracted region and the feature type.

In FIG. 11, reference numeral 111 denotes an area extracting unit;
Reference numeral 112 denotes an extraction area re-input unit, 113 denotes a simultaneous occurrence matrix calculation unit, and 114 denotes a covariance matrix update unit.
Note that the co-occurrence matrix calculation unit 113 is not an essential component.

FIG. 11 is characterized in that, based on the result once output as the saliency map, a partial area determined to be highly salient by the area extracting unit 111 is extracted. That is, the feature of each part is extracted again by the area extracting unit 111 from the output saliency map, and the extracted image data is input again from the extracted area re-input unit 112. The co-occurrence matrix is calculated again in the co-occurrence matrix calculation unit 113 based on the re-input image data, and the covariance matrix is updated in the covariance matrix update unit 114, so that the KL transform is performed again, and the main eigenvector itself is recalculated. What you want to do.

By doing so, the more you use the system, that is, the more experience you have on the system, the more
It is possible to realize a saliency calculation device that can determine the saliency using an appropriate feature type.

As described above, according to the first embodiment, the types of features to be extracted can be changed based on the learning image data or the image data determined to have high saliency.
The saliency can be determined based on the type of feature appropriate for the image data. Therefore, by emulating such a learning effect with the degree of human experience and memory, it is possible to determine the saliency more similar to the human sense.

The saliency calculating device according to the present embodiment
It can be applied in various fields. For example, application to a mobile terminal or the like having a limited display area on a screen can be considered. FIG. 12 is a processing flowchart in a portable terminal using the saliency calculation device according to one embodiment of the present invention.

In general, when a large image is displayed on a small screen such as a portable terminal, the image is reduced and displayed. In addition, details are often not displayed. In order to solve such a display problem, a part of an image is often cut out and displayed. The purpose of the present embodiment is to determine the saliency of an input image and cut out and display a highly salient portion in the image, that is, an important partial region in the image.

In FIG. 12, first, the entire image data to be displayed is input (step S121), and the saliency is calculated for the entire image data to obtain a saliency map for the entire image data (step S122).
The saliency calculation method according to the present embodiment is used for calculating the saliency.

Next, since there is a physical restriction on the size that can be displayed on the portable terminal, the size of the image area that can be displayed is specified (step S123), and the window of the image area size is set for the entire display image data. Raster scanning is performed sequentially (step S124), and saliency in the window area is calculated each time the position of the window is changed (step S125). The saliency calculation method according to the present embodiment is also used for calculating the saliency.

Then, a window region determined to have the highest saliency is selected as a cutout region (step S).
126), a screen with the highest saliency, that is, an image of a portion determined to contain the most important information for the user is displayed on the screen of the portable terminal (step S1).
27).

By doing so, the image data determined to contain the most prominent information, ie, the part that can be determined to be an important part by a person, is displayed on the screen of the portable terminal. In addition, since the image data is not displayed in a reduced size, the problem of losing details of the image can be solved.

It is also conceivable to apply an abnormality monitoring system using a TV camera to unmanned operation. FIG. 13 is a processing flowchart in an abnormality monitoring system using a saliency calculation device according to another embodiment of the present invention.

In FIG. 13, the input medium for inputting an image is a television camera. Image data to be monitored is input from the camera as a constantly moving image or as a still image at fixed time intervals (step S131). Then, the saliency of the image data input by the camera is calculated (step S132). That is, for a moving image, the saliency of image data at regular time intervals,
The saliency of each still image is calculated. For the calculation of the saliency, the saliency calculation method according to the present embodiment is used.

Then, by detecting a case where the saliency suddenly increases or suddenly increases (step S133: Y
es), it is determined that an abnormality has occurred (step S1)
34). For example, if a place where no person enters or exits is always monitored and saliency is calculated, the saliency is calculated to be very high when an image of a person or another living thing is input. Therefore, even if a person does not always look into the surveillance camera at all times, it is possible to easily monitor the entry of a person, a living thing, or the like by automatic judgment based on the quantitative value of the saliency.

Further, it is conceivable that the present invention is applied to a screen creation support device for assisting in enhancing advertisements, such as arranging information desired by the user in the design of a web screen or the like so as to increase the saliency. FIG. 14 is a flowchart of a process performed by the alert system on the web screen using the saliency calculating device according to another embodiment of the present invention.

In FIG. 14, first, an initial design screen is input as a saliency calculation target image (step S141).
Then, the saliency of the target image is calculated to create a saliency map (step S142). For the calculation of the saliency, the saliency calculation method according to the present embodiment is used.

Next, an object for displaying information desired by the user is specified (step S143), and the saliency of the area occupied by the object in the image is calculated (step S144). For the calculation of the saliency, the saliency calculation method according to the present embodiment is used.

If the saliency of the object area is larger than a predetermined threshold value T determined in advance, it is determined that the object area has sufficient saliency in the screen display, and the process is terminated (step S1). S145: Yes). Conversely, if the threshold value T is smaller than the predetermined threshold value T (step S145: No), it is determined that the threshold value does not have sufficient saliency, and the design of the target screen is changed. (Step S146), similar processing is repeated. Needless to say, such a design change is a change in the color or arrangement of the object.

The recording medium storing the program for realizing the saliency calculating device according to the embodiment of the present invention is:
As shown in the example of the recording medium shown in FIG.
152-1 and floppy (registered trademark) disk 152-
2 as well as other storage devices 151 provided at the end of the communication line and the computer 15.
3 or a recording medium 154 such as a RAM. When the program is executed, the program is loaded and executed on the main memory.

The recording medium on which the secondary feature vector or the like generated by the saliency calculating device according to the embodiment of the present invention is recorded is also a CD-ROM 152- as shown in the example of the recording medium shown in FIG. 1 or floppy disk 15
Not only a portable recording medium 152 such as 2-2, but also any other storage device 151 provided at the end of a communication line, or a recording medium 154 such as a hard disk or a RAM of a computer 153 may be used, for example, according to the present invention. It is read by the computer 153 when using the saliency calculation device.

[0073]

As described above, according to the saliency calculating apparatus according to the present invention, it is possible to change the type of feature to be extracted based on the learning image data or the image data determined to have high saliency. The saliency can be determined based on the type of feature that matches the data. Therefore, by emulating such a learning effect with the degree of human experience and memory, it is possible to determine the saliency more similar to the human sense.

[Brief description of the drawings]

FIG. 1 is a diagram showing an example of saliency judgment

FIG. 2 is a processing flowchart in a conventional saliency calculation method.

FIG. 3 is an illustration of an example of saliency judgment

FIG. 4 is an illustration of an example of saliency judgment

FIG. 5 is a basic configuration diagram of a saliency calculating device according to the embodiment of the present invention;

FIG. 6 is an explanatory diagram of a saliency calculation process in the saliency calculation device according to the embodiment of the present invention;

FIG. 7 is an explanatory diagram of pattern recognition in the saliency calculation device according to the embodiment of the present invention;

FIG. 8 is an exemplary view showing an output display in the saliency calculating device according to the embodiment of the present invention;

FIG. 9 is an exemplary view showing an output display in the saliency calculating device according to the embodiment of the present invention;

FIG. 10 is a flowchart of a process in the saliency calculating device according to the embodiment of the present invention;

FIG. 11 is a configuration diagram of a saliency calculation device according to the embodiment of the present invention.

FIG. 12 is a flowchart of a process in the saliency calculation device according to the embodiment of the present invention;

FIG. 13 is a flowchart of a process in a saliency calculating apparatus according to another embodiment of the present invention.

FIG. 14 is a flowchart of a process in a saliency calculating device according to another embodiment of the present invention.

FIG. 15 is an exemplary diagram of a recording medium.

[Explanation of symbols]

REFERENCE SIGNS LIST 51 image input unit 52 feature extraction unit 53 saliency calculation unit 54 output display unit 61 vector of image pattern 62 feature vector 63 secondary feature vector 64 difference of square norm 111 area extraction unit 112 extraction area re-input unit 113 simultaneous occurrence matrix calculation Unit 114 covariance matrix calculation unit 151 storage device at line destination 152 portable recording medium such as CD-ROM or floppy disk 152-1 CD-ROM 152-2 floppy disk 153 computer 154 recording medium such as RAM / hard disk on computer 521 Learning image group input unit 522 Feature type determination unit

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Makoto Goto 4-1-1, Kamidadanaka, Nakahara-ku, Kawasaki-shi, Kanagawa F-term in Fujitsu Limited (Reference) 5L096 CA24 FA19 FA28 FA33 FA81 GA17 HA09 JA05

Claims (6)

[Claims] 1. An image input unit for inputting first image data to be subjected to saliency calculation, a feature extraction unit for extracting features of each unit in the first image data, and a feature of each of the extracted units A saliency calculation device including a saliency calculation unit that quantifies saliency on the basis of an output display unit that displays and outputs the calculated saliency, wherein a plurality of second image data to be learned are input. A learning image group input unit that performs learning based on a plurality of second image data, and a feature type determining unit that determines and determines a feature type in advance. A saliency calculation apparatus, wherein the type of the feature is input to the feature extraction unit, and the feature extraction unit extracts a feature of each unit based on the type of the feature and the first image data. 2. The image processing apparatus according to claim 1, further comprising: an area extracting unit configured to extract third image data that is an area with high saliency from the first image data based on the saliency output from the output display unit. In place of the learning image group input unit, the learning image group input unit includes an extraction area re-input unit that inputs the third image data extracted by the area extraction unit, and the feature type determination unit uses the extraction type based on the third image data. 2. A feature type is learned, and the feature type to be extracted by the feature extraction unit is determined again.
The saliency calculation device according to the above. 3. The saliency calculation device according to claim 1, wherein the feature type determination unit uses KL conversion. 4. The saliency calculation device according to claim 1, wherein the saliency calculation unit uses KL conversion. 5. A step of inputting first image data to be subjected to saliency calculation, a step of extracting features of each part in the first image data, and a salient feature based on the extracted features of each part. A plurality of second image data to be learned, including a step of quantifying the degree of saliency, and a step of displaying and outputting the calculated saliency. And learning the type of the feature in advance based on the image data of (a), and determining the type of the feature based on the determined type of the feature and the first image data. A saliency calculation method characterized by the following. 6. A step of inputting first image data to be subjected to saliency calculation, a step of extracting features of each part in the first image data, and a salient feature based on the extracted features of each part. A computer-readable recording medium storing a program to be executed by a computer for realizing a saliency calculation method including a step of quantifying the saliency and a step of displaying and outputting the calculated saliency, which is to be learned. The method further includes a step of inputting a plurality of second image data and a step of learning and determining a type of a feature in advance based on the plurality of the second image data, and determining the type of the determined feature. A computer-readable program storing a program to be executed by a computer, wherein a feature of each unit is extracted based on the first image data. Recording medium that can be taken.