JP2004246478A - Image search device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image search device capable of reducing processing load. <P>SOLUTION: In this image search device for searching an image data part that is an object of search from object image data that are an object of processing, a control part 11 specifies search area candidates from the object image data according to a predetermined rule, extracts a plurality of partial search area candidates from the search area candidates according to a predetermined partial search area extraction rule, calculates the respective image characteristic quantity by use of the respective image data part defined by each partial search area candidate, counts the number of image characteristic quantities satisfying a predetermined characteristic quantity condition among the image characteristic quantities, determines, when the number satisfies a predetermined number condition, the corresponding search area candidate as a search area, and performs the processing of searching the object of search for the determined search area. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an image search device that searches for a specific image portion such as a face portion from image data such as a photograph.
[0002]
[Prior art]
In recent years, it has been considered that a specific target object included in a photograph or the like, for example, a portion such as a human face is specified, and a predetermined process is performed based on the specified portion. As an example, a face portion of each person is detected from a photographed photograph, and only the face portion is burned, or a human face portion is detected from a video being captured and subjected to face authentication processing, And so on.
[0003]
In a conventional apparatus for recognizing an object such as a face image, image data is divided into search areas according to a predetermined rule, and it is determined whether or not the search object is included in all of the search areas thus obtained. At this time, in order to cope with a case where it is difficult to recognize the target object depending on the imaging state (tilt, size, illumination state, and the like) of the target object, a process of adapting the imaging state to a predetermined imaging state (reference state) is performed. Do.
[0004]
Conventionally, in this processing, specifically, a neural network is learned using learning image data taken while changing the imaging state, and the processing is performed using the learned neural network. It has been considered to detect how much the imaging state deviates from the reference state and perform image processing to correct the deviation.
[0005]
As an example of a process of detecting a desired pattern from target image data, a method such as a kernel nonlinear subspace method disclosed in Patent Document 1 is known.
[0006]
[Patent Document 1]
JP 2001-90274 A
[Problems to be solved by the invention]
However, generally, processing for detecting a desired search target from target image data and processing for adapting to a reference state is a load-intensive processing. In the above-described conventional apparatus, each processing is performed for each search area, and thus the processing load increases.
[0008]
The present invention has been made in view of the above circumstances, and has as its object to provide an image search device capable of reducing the processing load.
[0009]
[Means for Solving the Problems]
The present invention for solving the problem of the above-mentioned conventional example is an image search apparatus for searching an image data portion of a search target from within target image data to be processed, wherein the target image is searched for in accordance with a predetermined rule. Search area identification means for identifying search area candidates from data; partial search area extraction means for extracting a plurality of partial search area candidates from the search area candidates according to a predetermined partial search area extraction rule; Using each of the image data portions defined in the above, each image feature amount is calculated, the number of image feature amounts satisfying a predetermined feature amount condition among each image feature amount is counted, and the number is determined to be a predetermined number. Means for determining a corresponding search area candidate as a search area when the number condition is satisfied, and searching for a search target in the determined search area. It is characterized by performing.
[0010]
Here, the partial search area extracting means may extract the partial search area candidate from the search area candidate while allowing partial overlap. Further, the search processing means includes a conversion method and a conversion amount to be applied to each of the image data portions included in the determined at least one search area with reference to a conversion database acquired in advance by learning. A conversion means for acquiring a conversion condition, and performing a conversion based on the acquired conversion condition at least once for the corresponding image data portion, and a search database learned and acquired using the image data to be searched in the reference state. And searching means for determining whether a search target is included in at least one of the converted image data portions.
[0011]
Further, the present invention for solving the problems of the above-described conventional example is an image search method for searching an image data portion to be searched from within target image data to be processed, and the method according to a predetermined rule. Identifying a search area candidate from the target image data; extracting a plurality of partial search area candidates from the search area candidate in accordance with a predetermined partial search area extraction rule; and an image defined by the partial search area candidates Using each of the data portions, the respective image feature amounts are calculated, the number of image feature amounts that satisfy a predetermined feature amount condition among the image feature amounts is counted, and the number satisfies the predetermined number condition. Determining a corresponding search area candidate as a search area, and performing a process of searching for a search target with respect to the determined search area. That.
[0012]
Further, the present invention for solving the problem of the above-mentioned conventional example is an image search program for searching an image data portion to be searched from within target image data to be processed. A step of specifying a search area candidate from the target image data according to a rule, a step of extracting a plurality of partial search area candidates from the search area candidate according to a predetermined partial search area extraction rule, and a step of defining each of the partial search area candidates Using each of the image data portions to be calculated, the respective image feature amounts are calculated, the number of image feature amounts satisfying a predetermined feature amount condition among the image feature amounts is counted, and the number is determined by a predetermined number condition. Is satisfied, a procedure of determining a corresponding search area candidate as a search area, and a process of searching for a search target in the determined search area are performed. It is characterized in that to execute the order, the.
[0013]
BEST MODE FOR CARRYING OUT THE INVENTION
[Basic configuration]
An embodiment of the present invention will be described with reference to the drawings. As shown in FIG. 1, the image search device according to the embodiment of the present invention includes a control unit 11, a storage unit 12, a database unit 13, a display unit 14, an operation unit 15, an external storage unit 16, , And is realized using a general computer configured. This computer may be incorporated in another product, for example, a camera.
[0014]
The control unit 11 operates in accordance with a program stored in the storage unit 12, and includes a search area defining process for defining at least one search area in the target image data to be processed, and a reference state. A conversion process for conversion, a search process for detecting a search area including a search target, and a predetermined process using a search result are executed. The specific processing contents of these control units 11 will be described later in detail.
[0015]
The storage unit 12 stores software executed by the control unit 11. The storage unit 12 also operates as a work memory for holding various data required by the control unit 11 during the processing. Specifically, the storage unit 12 can be realized as a storage medium such as a hard disk, a semiconductor memory, or a combination thereof.
[0016]
The database unit 13 is a database including a conversion database 13a used in the first conversion process of the control unit 11 and a search database 13b used in the search process, as described later. The database unit 13 is specifically a storage medium such as a hard disk, and the storage unit 12 may also serve as the database unit 13, but is separately illustrated here for the sake of explanation.
[0017]
The display unit 14 is, for example, a display device or a printer device, and displays information and the like in accordance with an instruction input from the control unit 11. The operation unit 15 is, for example, a keyboard, a mouse, or the like, and accepts a user operation and outputs the content of the operation to the control unit 11.
[0018]
The external storage unit 16 reads a program or data from a computer-readable removable medium (a type of storage medium) such as a CD-ROM or a DVD-ROM, and outputs the read program or data to the control unit 11. Is performed. The program according to the present embodiment can be distributed by being stored in a portable storage medium such as a CD-ROM, and is copied to the storage unit 12 using the external storage unit 16 and used. The program according to the present embodiment may be copied not only from such a storage medium but also from a server on a network to the storage unit 12 via a communication unit (not shown).
[0019]
[Process of Control Unit 11]
Here, the content of the process of the control unit 11 will be specifically described. In the present embodiment, image data (target image data) to be subjected to each processing is externally input via the external storage unit 16 or a communication unit (not shown) and stored in the storage unit 12. Here, the number of target image data may be one or plural. When the user operates the operation unit 15 to instruct the control unit 11 to perform a process of searching for a search target for specific target image data (instruction to start processing), the control unit 11 The process shown in FIG. 2 is started.
[0020]
The control unit 11 defines a search area for each reduced-converted target image data while sequentially reducing and converting the target image data, performs a conversion process to a reference state and a search process for each search region, and performs each reduction process. Map data is generated that indicates which part of the converted target image data contains the search target.
[0021]
Specifically, the control unit 11 first sets the reduction rate S to a minimum reduction rate (for example, 1 time, that is, does not reduce) (S1), and reduces the target image data at the reduction rate S (S2). Then, an area of the map data having a size equal to the size of the reduced target image data is secured in the storage unit 12, the value of the area is set to “false”, and the map data is initialized. (S3). For example, if the reduced target image data is image data of 1000 × 1000 pixels, an area for 1000 × 1000 bits is secured and each bit value is initialized to “0”.
[0022]
Next, the control unit 11 performs a process of determining at least one search area for the reduced target image data (S4). This search area determination processing will be described later in detail. Then, one of the search areas is selected (S5), and a conversion process is performed on the search area (S6). This conversion processing will also be described later in detail.
[0023]
The control unit 11 executes a process (search process) of determining whether or not a search target is included in the image data portion included in the search area after the conversion process (S7), and is determined to be included. If (Yes), the value of the area on the map data corresponding to the search area after the conversion processing is set to “true” (S8). Then, it is checked whether or not there is a search area that has not been selected (S9). If there is a search area that has not been selected (if Yes), the process returns to step S5 and one of the search areas that has not been selected is selected. To continue the process (A).
[0024]
On the other hand, when it is determined that the search target is not included as a result of the search process in the process S7 (No), the process directly proceeds to the process S9. In addition, if there is no unselected search area in the processing S9, that is, if the conversion processing and the search processing have been completed for all the search areas (if No), the currently set reduction rate S is set in advance. (S10), and if not (No), adjust so as to increase the reduction ratio S (S11), and return to the process S2 to process. (B). Here, the process S11 of adjusting the reduction rate S to a large value may be, for example, a process of increasing the reduction rate S at a predetermined ratio, or by multiplying the magnification which is the reduction rate S by a predetermined power factor ΔS to obtain S = A new reduction ratio S may be determined as S × ΔS.
[0025]
If the currently set reduction ratio S in the process S10 exceeds the predetermined maximum reduction ratio (Yes), the original reduction ratio is determined based on the map data corresponding to the target image data at each reduction ratio. A region including the search target is defined in the target image data (before reduction) (S12), and the process ends.
[0026]
[Search area determination processing]
Next, a process in which the control unit 11 determines a search area (search area determination processing) will be described. As illustrated in FIG. 3, the control unit 11 starts from the coordinates (for example, X = 0, Y = 0) of the upper left corner of the reduced target image data and determines a predetermined size L × L ′ (for example, 27 × 27). The rectangular area of × 27) is defined as a search area candidate (S20). In the following, for the sake of explanation, the coordinate system on this search area candidate is set as follows. That is, the upper left end is assumed to be η = 0, ξ = 0, the rightward direction is the positive η-axis direction, and the downward direction is the positive ξ-axis direction.
[0027]
The control unit 11 secures a storage area for the counter in the storage unit 12, and resets the counter to “0” (S21). Further, the control unit 11 initializes the coordinates (η, ξ) of the extraction start point of the partial search area to (0, 0). Next, the control unit 11 extracts n × n ′ (n <L, n ′ <L ′, for example, 9 × 9) partial search area candidates with (η, ξ) at the upper left corner from the search area candidates ( S22) For the image data portion determined by the partial search area candidate, a feature amount, for example, entropy e, which is predetermined according to the property of the search target is calculated (S23).
[0028]
Then, it is determined whether or not the calculated entropy e is larger than a predetermined threshold value eth1 (S24). If the calculated entropy e is larger (Yes), the counter is incremented (S25), and the partial search area candidate is determined. Is moved in the X-axis direction by a predetermined movement amount d (for example, three pixels, that is, partial overlap between partial search area candidates may be allowed) (that is, η = η + d; S26). The control unit 11 checks whether the new partial search area candidate has deviated from the right end of the search area candidate, that is, whether η + n> L has been satisfied (S27). With “0” and ξ = ξ + d, a partial search area candidate having a new (η, ξ) at the upper left corner is extracted (S28). The control unit 11 further checks whether or not the new partial search area candidate has deviated from the lower end of the search area candidate, that is, whether or not ξ + n ′> L ′ (S29). It is checked whether the value of the counter is equal to or greater than a predetermined threshold value TH (S30). If the value of the counter is equal to or larger than the threshold value TH (if Yes), the search area candidate being processed is determined as the search area (S31), and it is checked whether the next search area candidate should be defined (S31). It is checked whether there is still a definable search area candidate on the target image data; S32). If the search area candidate is to be defined, that is, if there is still a definable search area candidate (Yes), the process returns to step S20 to start. The coordinates of the point are moved by ΔX in the X direction, and when reaching the end in the X direction, X is set to “0” and moved by ΔY in the Y direction to define and process the next L × L ′ search area. Continue.
[0029]
In this process S32, if there is no search area candidate to be defined, that is, if there is still no definable search area candidate (No), the process ends.
[0030]
Further, in steps S27 and S29, if the respective values do not deviate from each other (if No in each case), the process returns to step S22 to continue the processing. If the value of the counter is smaller than the predetermined threshold value TH in step S30 (if No), the process directly proceeds to step S32 without setting the search area candidate being processed as the search area. Further, in the process S24, if the entropy e is not larger than the predetermined threshold value eth1 (if No), the process shifts to the process S26.
[0031]
Further, here, a condition of whether or not entropy e is greater than a predetermined threshold value eth1 is used as the feature amount condition, but entropy e is sandwiched between two predetermined threshold values eth1 and eth2. It may be set as a feature amount condition whether or not it is within a range to be performed.
[0032]
Here, the specific contents of the processing S23 in which the control unit 11 calculates the entropy e will be described. The control unit 11 generates a histogram of the luminance of the pixels included in the image data portion determined by the partial search area candidate subjected to the processing, further complements this histogram by, for example, linear approximation, and the like. Is calculated as entropy e. The reason why the histogram is complemented in this way is that the histogram is calculated as a discrete amount while the entropy value is originally a continuous amount. Therefore, the calculated luminance histogram is approximately represented by a continuous function, and the sum of the approximate continuous function as an integral is defined as a histogram.
[0033]
Note that the histogram may be complemented not by linear approximation, but by quadratic or higher approximation by interpolating the values of the histogram by a predetermined interpolation method.
[0034]
In this way, the relative relationship between the peak positions of the histogram is not considered in the mere summation, but the relative distance between the peak positions contributes to the integration result by integrating the approximate functions. As a result, the entropy profile is made highly accurate.
[0035]
Here, the moving amount of the start point is set to ΔX and ΔY in the width direction and the height direction, respectively. However, these moving amounts are different from ΔX and ΔY at 1 time in accordance with the reduction ratio S of the target image data. ΔX / S and ΔY / S.
[0036]
In the present embodiment, for each of the partial search area candidates included in the search area candidates, a feature amount such as entropy calculated from an image data portion determined by each partial search area candidate is determined by a predetermined condition. It is determined whether or not to determine the search area candidate as a search area depending on how many satisfy the condition.
[0037]
For example, in the case of a face image, the entropy is relatively high in the eyes, nose, contour, etc., whereas the entropy is relatively low in the face, forehead, cheeks, etc. When the feature amount is captured in the entire search area candidate, the entropy is reduced on average, and it is considered that it is difficult to capture the face part. That is, when there is a portion where the feature amount is high and a portion where the feature amount is not in the region, the feature amount is averaged over the entire region, thereby preventing the identification by the feature amount from becoming difficult.
[0038]
As described above, according to the present embodiment, before the conversion processing and the search processing, it is accurately determined whether or not the search target may be included, and the conversion processing and the search for the part having no possibility are performed. Since processing is not performed, the processing load is reduced.
[0039]
[Conversion processing]
Next, the conversion process will be described. One of the characteristic features of the present embodiment is that this conversion process is performed in stages, and in each stage, a conversion corresponding to one conversion degree of freedom is performed. The control unit 11 according to the present embodiment calculates predetermined feature amount vector information based on the image data portion included in the search area. Here, the feature amount vector information is a vector amount including a plurality of feature amount elements selected according to the property of the search target.
[0040]
In the present embodiment, the description will be made assuming that the control unit 11 specifies the conversion by the kernel nonlinear subspace method using the feature amount vector information and the feature amount vector information stored in the conversion database 13a.
[0041]
[Contents of Conversion Database 13a]
This kernel non-linear subspace method is widely known as a method of classifying data into some category, and therefore detailed description is omitted. Each of the plurality of subspaces Ω included in the space F is recognized as a category to which data is classified, and feature vector information (for example, Φ) in the space F created based on the data to be classified is defined. Each of the subspaces Ω is projected (the projection result is assumed to be φ, for example), and the subspace Ω at which the distance E between the feature vector information Φ before the projection and the feature vector information φ after the projection becomes the smallest (temporarily, This is a method of detecting the tangent subspace) and determining that the data to be classified belongs to the category represented by the subspace Ω.
[0042]
Therefore, in the learning stage, a non-linear mapping (mapping to space F, that is, mapping to space F, At least one of the parameters included in the kernel function) and the hyperplane separating the subspaces Ω corresponding to the respective categories is adjusted.
[0043]
In the present embodiment, the conversion database 13a is formed to determine a method (type and amount of conversion) of converting the search target into the reference state. That is, the conversion database 13a is learned and acquired so that the type of conversion to be performed and the amount (category) of conversion to be performed on the image data for which it is unknown whether or not the image data is in the reference state. In the present embodiment, the conversion database 13a is created for each type of conversion (degree of freedom) to be performed on the image, such as rotation, translation, and size change of the image. The conversion database 13a corresponding to each degree of freedom of conversion is obtained by learning the conversion amount of the corresponding conversion as a category.
[0044]
In order to acquire the learning, in the learning process of the conversion database 13a of the present embodiment, a learning sample is generated as follows. That is, a plurality of examples of image data to be searched in a predetermined reference state are prepared, and for each example of image data, conversion is performed with a different conversion amount for each degree of freedom. And generating a plurality of converted image data. The converted image data generated for each degree of freedom is used as a learning sample for each degree of freedom. When a face is specifically targeted for search, a plurality of image data of a face in a predetermined reference state (predetermined shooting condition / posture) are prepared as an example, and the degree of freedom of the conversion of each image data is set to, for example, rotation / rotation. For each degree of freedom such as parallel movement and size change, if it is a rotation, converted image data rotated at an angle of 5 degrees in the range from -180 to 180 degrees is used as a learning sample for the degree of freedom of rotation. . In the case of parallel movement, a plurality of converted image data, which are moved vertically and horizontally by 5 pixels each, are used as learning samples for the degree of freedom of the parallel movement. Note that these learning samples are generated by moving the area of the reference state by 5 pixels from the image data of the area wider than the reference state while including the degree of freedom of conversion such as movement.
[0045]
In this manner, for each of the plurality of image data examples, a plurality of image data items each of which is subjected to a plurality of conversions for each degree of freedom are generated, and information indicating the type of conversion performed on each image data item (the amount of conversion amount) Etc.).
[0046]
Here, in order to obtain image data that has been subjected to conversion with different conversion amounts from each other, the image data obtained by performing each conversion while changing the conversion amount by a predetermined step (for example, 5 degrees in terms of rotation) is used as a learning sample. Although they are included, the conversion may be performed while determining the conversion amount by random numbers, and may be included in the learning sample, instead of being changed by predetermined steps.
[0047]
Next, the conversion database 13a corresponding to each degree of freedom is trained using a learning sample for each degree of freedom.
[0048]
[Operation of conversion processing]
The control unit 11 performs the following conversion process on each of the search areas defined by the search area definition processing using the conversion databases 13a learned in this manner. That is, the image data portion (for example, which can be regarded as a vector value as a column of pixel values) included in the search area targeted for processing is converted into feature amount vector information in the space F (for each conversion database 13a, that is, (A set of feature amounts defined for each degree of freedom of the transformation), and the mapping is projected to each subspace Ω. Then, a conversion amount that minimizes the distance E between the feature amount vector information before projection and the feature amount vector information after projection is determined. Further, the control unit 11 calculates the square value L of the distance E, and stores this as an error in the storage unit 12.
[0049]
Here, the conversion amount is determined for each degree of freedom based on each conversion database 13a, but the control unit 11 determines one of the conversion amounts corresponding to each degree of freedom under a predetermined condition (for example, corresponding to each conversion amount). And the conversion corresponding to the selected degree of freedom is converted by the selected amount of conversion.
[0050]
In other words, from the image data portion included in the search area, information obtained by learning in each conversion database 13a corresponding to each degree of freedom is used. , And the error is Lr. With respect to the degree of freedom of the parallel movement, the pixel approaches the reference state by conversion of 5 pixels to the left, and information such as the error Lp is obtained. Choose a transformation with more degrees of freedom. For example, in the case of the above example, if Lr <Lp, a 10-degree rotation conversion is performed on the search area to define a new search area. Then, the image data portion included in the new search area is further mapped to feature amount vector information in the space F, and the mapping is further projected to each subspace Ω. Then, the process is repeated from the process of determining the conversion amount that minimizes the distance E between the feature amount vector information before projection and the feature amount vector information after projection.
[0051]
If the conversion amounts corresponding to the respective degrees of freedom are all “0” (that is, no conversion), the process is terminated at that stage. The conversion process is performed on any of the unprocessed search areas as a processing target.
[0052]
Here, the image data portion included in the defined search area in the target image data is used as it is, but a process of reducing the resolution of the image data portion is performed to obtain coarse-grained data, and the coarse-grained data is obtained. The conversion process may be performed using the data. In this case, each conversion database 13a is learned by using a learning sample corresponding to the coarse-grained data.
[0053]
In addition, the control unit 11 may perform processing such as calculation of feature amount vector information, mapping to a subspace, evaluation of distance, and evaluation of error in parallel, instead of sequentially for each degree of freedom.
[0054]
Further, although the case where the kernel nonlinear subspace method is used has been described as an example, other methods such as an auto encoder may be used as long as data classification and error evaluation at the time of classification are possible.
[0055]
[Search process]
Next, the search processing will be described. In this search processing, it is determined whether or not a search target is included in the image data portions included in each of the search areas for which the conversion processing has been completed, using the search database 13b. As a specific example of the search processing, there is a method disclosed in JP-A-2002-329188. Next, the outline will be described.
[0056]
[Study course of search database]
The search database 13b is formed by learning a neural network by using, as a learning sample, an example of image data to be searched in a reference state. That is, the control unit 11 receives a plurality of learning samples, calculates a set of feature amounts (feature amount vectors) preliminarily selected according to the properties of the search target for each of the samples, and generates learning data. I do. Next, a grid space map is formed by SOM (self-organizing map) on the M × M ′ grid space stored in the storage unit 12 using the learning data. That is, the control unit 11 calculates the distance between the input feature amount vector as the learning data and the weight vector assigned to each lattice by a predetermined measure (for example, a Euclidean measure), and determines that the distance is minimum. (A best matching node) c is detected. Then, for the plurality of grids near the best matching node, the weight vectors are updated using the input feature amount vectors. By repeating this process, a lattice space map is formed on the storage unit 12, and the best matching nodes for the feature amount vectors similar to each other form a continuous area. That is, in this grid space, a non-linear projection from a feature vector, which is a multi-dimensional input signal, to a two-dimensional map is formed while maintaining the phase, and by updating the weights, the characteristic part of the data is organized, As a result of the learning, a grid responding to similar data is present in close proximity.
[0057]
When the learning based on each learning data is completed, the control unit 11 classifies each grid of the grid space map into a category. This classification can be performed based on, for example, the distance between the respective lattices (the distance between the weight vectors associated with the respective lattices), and a category of a lattice group that responds to image data similar to the search target (search target category). And the other categories of the grid group (non-search target categories).
[0058]
[Operation of search processing]
The control unit 11 secures an area for storing map data of the same size as the target image data in the storage unit 12, and initializes the value of the area to “false”.
[0059]
The control unit 11 calculates a predetermined feature vector based on the image data portion in the search area for which the conversion process has been completed, using the search database 13b acquired by learning. Then, the distance between the calculated feature vector and the weight vector associated with each grid in the search database 13b is determined, and the grid (best matching node) with the minimum distance to the feature vector is specified. If it belongs to the search target category, it is determined that the search target includes the search target.If the specified lattice belongs to the non-search target category, it is determined that the search target does not contain the search target. to decide.
[0060]
[Operation of Control Unit 11]
As described above, while sequentially reducing the target image data to be searched, the control unit 11 according to the present embodiment extracts an area for performing a search process from each of the reduced target image data, and After the image data is converted so as to be in the reference state, it is determined whether or not the search target is included in the image data in the search area after the conversion. That is, since the control unit 11 performs the process of setting the image data to be searched for as the reference state, the position of the search area before the conversion may be slightly shifted. In addition, there is no problem even if the reduction ratio is slightly deviated from the reference state.
[0061]
Therefore, in the related art, reduced image data in multiple stages reduced by 0.8 times is generated, and the search area is extracted while being shifted one pixel at a time. It is possible to reduce by 5 times, and when defining a search area, even when autonomously extracting an area satisfying a predetermined condition, ΔX and ΔY can be set to 6 pixels or the like. As a result, the number of patterns to be searched can be significantly reduced, and the load of processing for searching for a search target from a photograph or the like can be reduced.
[0062]
The control unit 11 generates map data representing an area determined to include the search target as search result information, and this map data corresponds to each of the target image data after being reduced at each reduction rate. Are generated. Therefore, an area including the search target is determined by integrally using the plurality of map data (each having the size of the reduced target image data).
[0063]
For example, each map data is enlarged at an enlargement ratio corresponding to the respective reduction ratio, compared with the size of the original target image data and compared, and an area ("true") common to all map data ( Regardless of the target image data at any reduction ratio, it may be determined that the search target is included in the area (the area determined to include the search target). Further, an area that is “true” in any one of the map data may be determined to be an area including the search target.
[0064]
[Other Modifications]
In the description so far, the conversion performed in the conversion processing includes two-dimensional rotation, translation, and scaling (the search area is scaled, and the image data portion inside the search area is changed to the original (before scaling) search area). However, in the case of a human face, for example, a three-dimensional rotation affected by the posture (adjusting downward or turning) may be included. . In this case, assuming an average three-dimensional model to be searched, a three-dimensional rotation conversion can be realized using a projection of the image data portion onto the average three-dimensional model.
[0065]
In the search process, even if the search target is, for example, a human face, the category may be further divided into categories, and conditions such as age, gender, and whether or not the mouth is open may be included.
[0066]
Further, in the flowchart of FIG. 2, the control unit 11 sequentially performs the processing at each reduction rate. However, since the processing at each reduction rate is independent of each other, the control unit 11 The processing may be performed in parallel.
[Brief description of the drawings]
FIG. 1 is a configuration block diagram of an image search device according to an embodiment of the present invention.
FIG. 2 is a flowchart illustrating an example of a process of a control unit 11;
FIG. 3 is a flowchart illustrating an example of a process of determining a search area by a control unit 11;
[Explanation of symbols]
11 control unit, 12 storage unit, 13 database unit, 14 display unit, 15 operation unit, 16 external storage unit.

Claims (5)

An image search apparatus that searches for image data portions to be searched from within target image data that has been processed,
Search area specifying means for specifying a search area candidate from the target image data according to a predetermined rule;
A partial search area extraction unit configured to extract a plurality of partial search area candidates from the search area candidates according to a predetermined partial search area extraction rule;
Using each of the image data portions defined by the respective partial search area candidates, the respective image feature amounts are calculated, and the number of image feature amounts satisfying a predetermined feature amount condition among the image feature amounts is counted. Means for determining a corresponding search area candidate as a search area when the number satisfies a predetermined number condition;
Including
An image search apparatus, which performs a process of searching for a search target in the determined search area. 2. The image search apparatus according to claim 1, wherein the partial search area extracting unit extracts the partial search area candidate from the search area candidate while allowing partial overlap. The image search device according to claim 1 or 2,
The search processing means includes a conversion method and a conversion amount to be applied to each of the image data portions included in the determined at least one search region with reference to a conversion database acquired in advance by learning. A conversion unit that obtains a conversion condition, and performs conversion based on the obtained conversion condition at least once for a corresponding image data portion;
A search means for referring to a search database acquired by learning using image data to be searched in the reference state, and determining whether or not a search object is included in at least one of the converted image data portions; and ,
An image search device comprising: An image search method for searching an image data portion of a search target from target image data that has been processed,
Identifying a search area candidate from the target image data according to a predetermined rule;
Extracting a plurality of partial search area candidates from the search area candidates according to a predetermined partial search area extraction rule;
Using each of the image data portions defined by the respective partial search area candidates, the respective image feature amounts are calculated, and the number of image feature amounts satisfying a predetermined feature amount condition among the image feature amounts is counted. Determining the corresponding search area candidate as a search area when the number satisfies a predetermined number condition;
Including
An image search method characterized by performing a process of searching for a search target in the determined search area. An image search program for searching for an image data portion to be searched from within the target image data that has been processed, the computer comprising:
A procedure for specifying a search area candidate from the target image data according to a predetermined rule,
Extracting a plurality of partial search area candidates from the search area candidates according to a predetermined partial search area extraction rule;
Using each of the image data portions defined by the respective partial search area candidates, the respective image feature amounts are calculated, and the number of image feature amounts satisfying a predetermined feature amount condition among the image feature amounts is counted. Determining a corresponding search area candidate as a search area when the number satisfies a predetermined number condition;
Performing a process of searching for a search target for the determined search area;
And an image search program.

Images