JP2006092151A - Device, method and program for detecting area - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an area detection technique for speeding up detection processing to detect an image area of a specified object from an input image by utilizing a partial space obtained from an image sample assembly. <P>SOLUTION: A device for detecting an area detects whether or not a face (specified object) is included in an image area cut out from the input image by utilizing a feature space based on the feature analysis of the image sample assembly of the face. This detection converts pixels in the image area into vectors and firstly projects it to one base vector among a plurality of base vectors which establish the feature space to calculate a distance from feature space (DFFS) and a distance in feature space (DIFS). Here, if the calculated DFFS and DIFS are not within prescribed ranges, the device determines that the face is not present in the image area and stops projection to other base vectors. As a result, the device can speed up the detection processing of the specified object. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an area detection technique for detecting an image area related to a specific object from an input image.

  A detection method for detecting a region including a specific object (specific object) from images having various backgrounds is disclosed in Patent Document 1, for example. In this detection method, a rectangular area (window) defined in an image is scanned and moved over the entire screen, and it is determined whether an image area sequentially cut out in the window includes a specific object. This method will be specifically described below.

  First, each pixel in the window is vectorized and projected onto the principal component space. Then, a distance to the principal component space (DFFS: Distance From Feature Space) and a distance in the principal component space (DIFS: Distance In Feature Space) are calculated, and whether or not the image area includes a specific object based on the calculation result. Determine whether. By projecting onto the principal component space as described above, an object can be extracted more robustly than the detection method using simple template matching.

  The above principal component space is a partial space configured so that a principal component analysis is performed on a group of image samples of a specific object collected in advance, and the mean square residual to the space for the sample group is minimized. Yes. Here, in general, the appearance of a specific object on an image varies depending on various factors, such as a face pattern that changes depending on an individual or facial expression if it is a human face. A multidimensional principal component space corresponding to each mode of variation is extracted.

JP-A-8-339445

  However, in the technique of the above-mentioned Patent Document 1, since each image region sequentially cut out in a window that scans the entire image is projected onto the entire multidimensional principal component space, the calculation time becomes long. There is.

  The present invention has been made in view of the above problems, and provides an area detection technique capable of speeding up a detection process for detecting an image area of a specific object from an input image using a partial space obtained from a set of image samples. The purpose is to do.

  In order to solve the above problems, the invention of claim 1 is an area detection device for detecting an image area related to a specific object from an input image, and (a) corresponds to a detection area included in the image. Extracting means for extracting an image region from the image, and (b) vectorizing specific numerical information regarding the image region extracted by the extracting means, and subspace obtained from the image sample set relating to the specific object Projection means for projecting to one of the plurality of base vectors to be stretched, and (c) determining whether the image area is an area including the specific object based on the result of projection by the projection means Determination means.

  According to a second aspect of the present invention, in the region detection device according to the first aspect of the invention, the partial space is an orthogonal complementary space of a feature space that represents a feature of the specific object.

  Further, the invention of claim 3 is the area detecting apparatus according to claim 2, wherein the projecting means is (b-1) when the number of dimensions of the orthogonal complement space is smaller than the number of dimensions of the feature space, Means for projecting to one basis vector relating to the orthogonal complement space;

  According to a fourth aspect of the present invention, in the region detection apparatus according to any one of the first to third aspects of the present invention, the projection means includes (b-2) assigning each sample of the image sample set to the plurality of bases. Means for projecting to each basis vector in the order of basis vectors having a large variance when projected onto each vector.

  According to a fifth aspect of the present invention, there is provided a program for causing a region detection device to function as the region detection device according to any one of the first to fourth aspects by being executed by a computer incorporated in the region detection device. It is.

  The invention according to claim 6 is an area detection method for detecting an image area related to a specific object from an input image, wherein (a) an image area corresponding to a detection area included in the image is defined as the image area. (B) vectorizing specific numerical information related to the image area extracted in the extraction step, and a plurality of base vectors extending a subspace obtained from the image sample set related to the specific object A projection step of projecting onto one of the basis vectors, and (c) a determination step of determining whether the image region is a region including the specific object based on a result of projection in the projection step. .

  According to the first to sixth aspects of the present invention, a plurality of base vectors that create a vector of specific numerical information related to an image region extracted from an image and span a partial space obtained from a set of image samples related to a specific object. Projection is performed on one of the basis vectors, and based on the projection result, it is determined whether or not the image region is a region including a specific object. As a result, it is possible to speed up the detection process for a specific object.

  In particular, according to the second aspect of the present invention, since the subspace is an orthogonal complementary space of the feature space that expresses the feature of the specific object, it is possible to speed up the determination process for the image region that does not include the specific object.

  According to the third aspect of the present invention, when the number of dimensions of the orthogonal complement space is smaller than the number of dimensions of the feature space, projection is performed on one basis vector related to the orthogonal complement space, so that detection processing can be performed at higher speed.

  Further, in the invention of claim 4, since each sample of the image sample set is projected onto each of the plurality of basis vectors, the basis vectors are projected in the order of the basis vectors in which the variance becomes large. Can be efficiently performed.

<First Embodiment>
<Main part configuration of area detection device>
FIG. 1 is a schematic diagram showing the main configuration of an area detection apparatus 1A according to the first embodiment of the present invention.

  The area detection device 1A is configured as a personal computer, for example, and can detect an image area related to a specific object such as a face from an input image (detailed later). This area detection device 1 </ b> A includes a processing unit 10 having a box shape, an operation unit 11, and a display unit 12.

  A drive 101 for inserting the optical disk 91 and a drive 102 for inserting the memory card 92 are provided on the front surface of the processing unit 10.

  The operation unit 11 includes a mouse 111 and a keyboard 112, and receives an operation input from the user to the area detection device 1A.

  The display unit 12 is composed of, for example, a CRT and functions as a display unit.

  FIG. 2 is a diagram illustrating functional blocks of the area detection device 1A.

  The processing unit 10 of the area detection apparatus 1A includes an input / output I / F 13 connected to the operation unit 11 and the display unit 12 and a control unit 14 connected to the input / output I / F 13 so as to be able to transmit. The processing unit 10 also includes a storage unit 15 that is connected to the control unit 14 so as to be able to transmit, an input / output I / F 16, and a face area search unit 17.

  The input / output I / F 13 is an interface for controlling transmission / reception of data between the operation unit 11, the display unit 12, and the control unit 14.

  The storage unit 15 is configured as a hard disk, for example, and stores an area detection program for detecting a face area.

  The input / output I / F 16 is an interface for inputting / outputting data to / from the optical disk 91 and the memory card 92 as recording media via the drives 101 and 102.

  The face area search unit 17 is a part for detecting a face area (candidate) from the input image.

  The control unit 14 includes a CPU 141 that functions as a computer and a memory 142, and is a part that performs overall control of the operation of the area detection device 1A. By executing the region detection program in the storage unit 15 by the control unit 14, the face region search unit 17 and the like can be controlled to perform face detection regarding the input image.

  Program data such as a face area detection program recorded on the optical disc 91 can be stored in the memory 142 of the control unit 14 via the input / output I / F 16. Thereby, the stored program can be reflected in the operation of the area detection apparatus 1A.

<About face area detection>
FIG. 3 is a diagram for explaining detection of a face area.

  First, an image is input to the area detection device 1A from the memory card 92 on which an image photographed by a digital camera, for example, is recorded via the input / output I / F 16 functioning as the image input unit 21.

  Then, the image area cutout unit 171 of the face area detection unit 17 cuts out an image area corresponding to the window WD to be raster scanned as indicated by an arrow SN from the input image GI (see FIG. 4).

  Next, the pre-processing unit 172 performs masking processing on the image region cut out by the image region cutting unit 171 and performs processing such as discrimination of a non-face region and normalization of luminance.

  The identification unit 173A of the face area search unit 17 vectorizes the pixels of the image area in the window, and projects the image area into the feature space representing the facial features to determine whether the face region is a face region. In the determination of the face area, an identification dictionary 174 in which a basis vector that spans a feature space (subspace) derived by statistical learning using a large number of face image samples and threshold values of DFFS / DIFS are stored is referred to. The

  The post-processing unit 22 integrates overlapping positions and sizes of face candidates in the input image detected by the face area search unit 17 to generate a final detection frame FR (see FIG. 4). . The final result in the post-processing unit 22 is displayed on the display unit 12.

<Operation of Area Detection Device 1A>
FIG. 5 is a flowchart showing the basic operation of the area detection apparatus 1A. This operation is performed by executing a region detection program in the storage unit 15 in the control unit 14.

  First, an image is input to the image input unit 21 by loading a memory card 92 storing an image whose face area is to be detected into the drive 102 and reading the image (step S1).

  Next, in step S2, the image area cutout unit 171 sets a window area corresponding to the face detection area in the input image.

  Specifically, as shown in FIG. 4, raster scanning is performed on a window WD having a size of, for example, 20 × 20 pixels included in the input image GI, and an image corresponding to the size of the window WD over the entire input image GI. The region is sequentially cut out from the input image GI.

  In step S3, the pre-processing unit 172 performs a masking process on the four corners of the image area extracted by the window WD in order to remove the influence of the background not related to the facial features.

  In step S4, the pre-processing unit 172 determines whether or not the image area processed in step S3 is a non-face area that is not a face. This non-face region discrimination is described in, for example, Japanese Patent Application Laid-Open No. 2003-22441 (“region extraction method, region extraction program, computer-readable recording medium recording the region extraction program, and region extraction device”). Using a discrimination method, a low-bright and dark image area or the like is determined as a non-face area. Here, the image area set as the non-face area is discarded.

  In step S5, the pre-processing unit 172 performs luminance normalization processing on the image area that has not been discarded as a non-face area by the operation in step S4. For example, plane fitting normalization for correcting the luminance gradient in the image region and histogram flattening processing for converting the luminance histogram so as to assign the same number of pixels to all luminance levels are performed.

  In step S6, the identification unit 173A vectorizes the luminance value (numerical information) of each pixel in the image area smoothed by the operation in step S5, and performs face area determination processing using the principal component space as the feature space. Do (details later).

  In step S7, it is determined whether scanning of the window WD has been completed for the entire input image GI. If the scan is completed, the process proceeds to step S8. If the scan is not completed, the process returns to step S2 in order to perform face detection for the next image area cut out by the window WD.

  In step S8, the post-processing unit 22 determines a detection frame for the face area candidate in which the face is detected in step S6. Specifically, assuming that the same face is detected in a plurality of face area candidates whose positions are close, these positions and sizes are averaged and integrated into one detection frame. For face area candidates that overlap each other, one detection frame is selected based on DFFS or DIFS, and the other is regarded as a false detection and deleted.

  In step S9, the detection frame determined in step S8 is displayed on the display unit 12 as a face area detection result, for example, as in the detection frame FR of FIG.

  FIG. 6 is a flowchart corresponding to the above step S6 and showing the determination operation in the feature space.

  In step S11, the luminance value of each pixel in the image area in which the luminance is normalized in step S5 is vectorized. Here, a vector is defined in which each pixel in the image area is arranged in the order of raster scanning, for example.

  In step S12, 1 is substituted into the basis number i of the basis vector. The basis vector number will be briefly described.

A principal component analysis is performed on the face image sample set, and a set of n basis vectors orthonormalized in the principal component space is defined as {e ₁ , e ₂ ,..., E _n }. For these basis vectors, the eigenvalues in the covariance matrix are sorted in descending order (descending order), that is, the samples of the face image sample set are vectorized and projected onto each of the n basis vectors in order of increasing variance. The base number is attached.

In step S13, the cumulative calculation of DIFS for the i-th basis vector e _i is performed.

For example, when the base number i = 1, sub _DIFS (1), which is a DIFS with respect to the e ₁ axis, is calculated as the square of l ₂ norm as shown in the following equation (1).

In this equation (1), (,) represents an inner product, and x ₀ represents an average vector. The same applies to the following equations.

For sub _DIFS (1) in the above equation (1), as shown in the conceptual diagram of FIG. 7, when the vector (xx ₀ ) is projected in the first principal component direction defined by the basis vector e ₁ . It represents the distance.

In step S14, it is determined whether the DIFS calculated in step S13 is within a predetermined range. For example, is compared with a threshold epsilon _DIFS, or greater than the threshold epsilon _DIFS is determined. If the DIFS is within the predetermined range, the process proceeds to step S15. If the DIFS is out of the range, the image area is determined to be a non-face area, and the process exits the subroutine and proceeds to step S7.

In step S15, the DFFS cumulative calculation for the i-th basis vector e _i is performed.

For example, in the case of the base number i = 1, sub _DFFS (1) which is DFFS with respect to the e ₁ axis is calculated as the square of l ₂ norm as shown in the following equation (2).

As shown in the conceptual diagram of FIG. 7, the sub _DFFS (1) of the above equation (2) is defined by the difference between the vector (x−x ₀ ) and the above sub _DIFS (1), that is, the basis vector e ₁ . It represents the distance to the first principal component.

In step S16, it is determined whether or not the DFFS calculated in step S15 is within a predetermined range. For example, it is compared with a threshold epsilon _DFFS, if this threshold value epsilon _DFFS smaller is judged. If DFFS is within the predetermined range, the process exits from this subroutine and proceeds to step S7. If it is out of the range, the process proceeds to step S17.

  In step S <b> 17, it is determined whether calculation for all n basis vectors spanning the principal component space has been completed. Here, if the calculation is completed, it is determined that no face is detected in the image area cut out in the window WD, and the process exits from this subroutine and proceeds to step S7. If not completed, the process proceeds to step S18.

  In step S18, (i + 1) is substituted for the base number i.

  By the operations in steps S13 to S16 described above, it is determined whether or not the image region is a region including a face based on the result of projection onto one base vector among a plurality of base vectors spanning the principal component space. If the DIFS exceeds a predetermined threshold, the face is not present, and if the DFFS is within the predetermined threshold, the face is present. It will be possible to improve efficiency.

  With regard to the above steps S13 and S15, a simple supplementary explanation will be given below regarding cumulative calculation of DIFS and DFFS up to a base number j (2 ≦ j <n).

A set of basis vectors corresponding to the basis numbers _{1 to j} is defined as {e ₁ , e ₂ ,..., E _j }, and a partial DIFS obtained by adding the respective DIFS and DFFS is defined as s _DIFS (j) and partial DFFS. _Is s _DFFS (j).

Here, s _DIFS (j + 1), which is a partial DIFS considering up to the base vector e _{j + 1} , is calculated by the following equation (3).

Then, the above equation (3) is calculated portion DIFS of s _DIFS (j + 1) in step S14, it is determined whether the threshold epsilon _DIFS greater is if the threshold epsilon _DIFS greater than, non-face It will be discarded as an area.

On the other hand, s _DFFS (j + 1), which is a partial DFFS considering the basis vector e _{j + 1} , is calculated by the following equation (4).

Then, the above equation (4) is calculated portion DFFS of s _DFFS (j + 1) in step S16, whether larger than the threshold epsilon _DFFS is determined, if the threshold epsilon _DFFS smaller than the face area It will be judged as.

  By the above-described operation of the area detection apparatus 1A, the DIFS and DFFS for each base vector extending the principal component space are sequentially calculated, and the calculation of other base vectors is terminated according to the calculation result, thereby speeding up the face detection process. it can. In addition, since the projection is performed in the order of basis vectors in which the variance of the image sample set is large, it is possible to efficiently determine whether or not it is a face region.

  In the area detection apparatus 1A, it is not essential to perform face detection using the principal component space of the face, and a non-face principal component space obtained from a non-face sample set may be used. .

Second Embodiment
The area detection apparatus 1B according to the second embodiment of the present invention has a similar configuration to the area detection apparatus 1A according to the first embodiment shown in FIGS. 1 and 2, but the identification unit shown in FIG. 3 is different. ing.

  That is, the identification unit 173B of the second embodiment performs face area detection described below.

<About face area detection>
If the number of dimensions of the principal component space as the feature space is higher than the number of dimensions of the orthogonal complement space as a result of the principal component analysis on the face sample set, the projection calculation to each basis vector for the orthogonal complement space, that is, By calculating DIFS and DFFS, the processing speed can be increased. By using this, the identification unit 173B of the region detection device 1B performs face detection. The orthogonal complement space will be specifically described.

  If the principal component space obtained from the face sample set is F and its orthogonal complement space is F⊥, the original space X of the feature vector is expressed by the following equation (5).

  When the two subspaces are expressed using basis vectors, the following equation (6) is obtained.

  That is, the principal component space F is defined by the basis vectors of the base numbers 1 to n, and the orthogonal complement space F⊥ is defined by the base vectors of the base numbers (n + 1) to m. Here, m is a number corresponding to the dimension of the original space X, and corresponds to, for example, the number obtained by subtracting the number of pixels in the mask area from the number of pixels constituting the image area in the window WD.

  Here, when n> (mn), that is, when the number of dimensions of the principal component space is larger than the number of dimensions of the orthogonal complement space, the projection calculation to the orthogonal complement space is performed by the projection calculation to the principal component space. The amount of calculation is smaller. In such a case, the identification unit 173B detects the face area using the orthogonal complement space.

  About operation | movement of the above area | region detection apparatus 1B, operation | movement similar to 1 A of area | region detection apparatuses of 1st Embodiment shown to the flowchart of FIG. 5 is performed. However, the operation described below is performed as the operation in step S6.

  FIG. 8 is a flowchart showing the determination operation in the feature space in the area detection device 1B. As a premise of this operation, it is assumed that the dimension of the principal component space is higher than the dimension of the orthogonal complement space.

  In step S21, the same operation as step S11 shown in the flowchart of FIG. 6 is performed.

  In step S22, n + 1 is substituted for the basis number i of the basis vector. As a result, it is possible to start projection calculation for the first basis in the orthogonal complement space, not in the principal component space.

  As in the first embodiment, the base vectors of the orthogonal complement space are sorted in order of increasing variance when the face samples are projected, and are assigned base numbers.

In step S23, DFFS is calculated for the i-th basis vectors e _i .

For example, in the case of the base number i = n + 1 corresponding to the first base of the orthogonal complement space, sub _DFFS (n + 1) which is a DFFS with respect to the en _{+ 1} axis is _expressed by the following equation (7). Is calculated as the square of the l ₂ norm.

In step S24, it is determined whether or not the DFFS calculated in step S23 is within a predetermined range. For example, it is compared with a threshold epsilon _DFFS, greater than the threshold epsilon _DFFS is determined. If DFFS is within the predetermined range, the process proceeds to step S25. If DFFS is out of the range, the process exits the subroutine as a non-face area and proceeds to step S7.

At step S25, the calculation of DIFS concerning the base vectors e _i to i-th.

For example, in the case of the base number i = n + 1, sub _DIFS (n + 1), which is a DIFS for the e _{n + 1} axis, is calculated as the square of l ₂ norm as in the following equation (8). .

In step S26, it is determined whether or not the DIFS calculated in step S25 is within a predetermined range. For example, it is compared with a threshold epsilon _DIFS, or the threshold epsilon _DIFS smaller is judged. If the DIFS is within the predetermined range, the process proceeds to step S27. If the DIFS is out of the range, the process exits the subroutine and proceeds to step S7.

  In step S27, it is determined whether the calculation for all (m−n) basis vectors spanning the orthogonal complement space is completed. Here, when the calculation is completed, it is determined that a face is detected in the image region cut out by the window WD, and the process exits from this subroutine and proceeds to step S7. If not completed, the process proceeds to step S28. .

  In step S28, (i + 1) is substituted for the base number i.

  As a result of the operations in steps S23 to S26 described above, when DIFS or DFFS exceeds a predetermined threshold, it is determined that no face exists, and the DIFS and DFFS operations relating to other basis vectors are terminated. It will be possible to plan.

  With respect to the above steps S23 and S25, a simple supplementary explanation will be given below regarding the calculation of DFFS and DIFS up to the base number j (n + 2 ≦ j <m).

A set of basis vectors corresponding to the basis numbers (n + 1) to _j is defined as {e _{n + 1} , e _{n + 2} ,..., E _j }, and a partial DFFS obtained by adding the respective DFFSs and DIFSs is defined as s _DFFS. (j), the partial DIFS is s _DIFS (j).

Here, s _DFFS (j + 1), which is a partial DFFS considering the basis vector e _{j + 1} , is calculated by the following equation (9).

Then, the above equation (9) is calculated portion DFFS of s _DFFS (j + 1) in step S24, whether larger than the threshold epsilon _DFFS is determined, if the threshold epsilon _DFFS smaller than the face area It will be judged as.

On the other hand, s _DIFS (j + 1), which is a partial DIFS considering the basis vector e _{j + 1} axis, is calculated by the following equation (10).

Then, the above equation (10) is calculated portion DIFS of s _DIFS (j + 1) in step S26, it is determined whether the threshold epsilon _DIFS greater is if the threshold epsilon _DIFS greater than, non-face It will be discarded as an area.

  By the above operation of the area detection apparatus 1B, the DIFS and DFFS for each base vector extending the orthogonal complement space are sequentially calculated, and the calculation of the other base vectors is terminated according to the calculation result. Speed can be increased. Furthermore, when the number of dimensions of the orthogonal complementary space is smaller than the number of dimensions of the principal component space, higher-speed processing can be performed by performing projection calculation on the orthogonal complementary space.

<Modification>
In each of the above embodiments, it is not essential to input an image from a recording medium such as a memory card, and an image may be input from a digital camera or a scanner connected to the area detection device.

  In each of the above embodiments, not only scanning the window for the input image, but also creating a plurality of images with different resolutions based on the input image, and sequentially scanning the window for these images. Anyway. By generating the image pyramid in this way, reliable face detection can be performed even when a face having a different size from the input image is detected.

  In each of the above embodiments, it is not essential to detect a face area by an information processing apparatus such as a personal computer, and the digital camera may be provided with a function for detecting a face area.

  In each of the above embodiments, it is not essential to detect a human face, and specific objects such as human organs, facial parts (for example, eyes and mouth), and automobiles may be detected.

  The feature space in each of the above embodiments is not necessarily generated using principal component analysis, and may be generated using discriminant analysis or the like.

  ◎ For the detection of the face area in each of the above embodiments, it is not essential to use the principal component space obtained from the face sample set, and the principal component space obtained from the pseudo face (non-face close to face) sample set is used. May be used. For example, when face detection is performed using a subspace method such as a graphic method or a projection distance method, two DFFSs up to a principal component space of a face and a principal component space of a pseudo face are compared. Thereby, after calculating the DFFS up to the principal component space of the face, the distance with the pseudo face space can be compared at high speed.

It is the schematic which shows the principal part structure of 1 A of area | region detection apparatuses which concern on 1st Embodiment of this invention. It is a figure which shows the functional block of 1 A of area | region detection apparatuses. It is a figure for demonstrating the detection of a face area | region. It is a figure for demonstrating extraction of the image area | region by the window WD. It is a flowchart which shows the basic operation | movement of the area | region detection apparatus 1A. It is a flowchart which shows the determination operation | movement in feature space. It is a conceptual diagram for demonstrating DIFS and DFFS. It is a flowchart which shows the determination operation | movement in the feature space in the area | region detection apparatus 1B which concerns on 2nd Embodiment of this invention.

Explanation of symbols

1A, 1B area detection device 12 display section 17 face area search section 171 image area cutout section 172 preprocessing section 173A, 173B identification section 174 identification dictionary 22 postprocessing section 91 memory card GI input image WD window

Claims (6)

An area detection device that detects an image area related to a specific object from an input image,
(a) extraction means for extracting an image area corresponding to a detection area included in the image from the image;
(b) Specific numerical information regarding the image region extracted by the extraction unit is vectorized, and projected onto one base vector among a plurality of base vectors extending a partial space obtained from the image sample set related to the specific object. Projection means to
(c) determination means for determining whether the image area is an area including the specific object based on the result of projection by the projection means;
An area detection apparatus comprising: The area detection apparatus according to claim 1,
The region detection device according to claim 1, wherein the partial space is an orthogonal complementary space of a feature space expressing the feature of the specific object. In the area detection device according to claim 2,
The projecting means includes
(b-1) means for projecting to one basis vector related to the orthogonal complement space when the dimension number of the orthogonal complement space is smaller than the dimension number of the feature space;
An area detection apparatus comprising: In the area | region detection apparatus in any one of Claim 1 thru | or 3,
The projecting means includes
(b-2) means for projecting each sample of the image sample set to each base vector in the order of base vectors having a large variance when each sample is projected onto each of the plurality of base vectors;
An area detection apparatus comprising:   An area detection program for causing an area detection device to function as the area detection device according to any one of claims 1 to 4 by being executed by a computer incorporated in the region detection device. An area detection method for detecting an image area related to a specific object from an input image,
(a) an extraction step for extracting an image region corresponding to a detection area included in the image from the image;
(b) Specific numerical information regarding the image region extracted in the extraction step is vectorized and projected onto one base vector among a plurality of base vectors spanning a partial space obtained from the image sample set related to the specific object. A projection process to
(c) a determination step of determining whether the image region is a region including the specific object based on the result of the projection in the projection step;
An area detection method comprising:

Images