KR101301301B1 - Face identification filter planning method and apparatus thereof - Google Patents

Abstract

PURPOSE: A face recognition filter design method and a device thereof are provided to facilitate the design of a filter with an improved face recognition capability by obtaining maximum sensitivity in variances for representing the face identity for different people and minimum sensitivity in matters unrelated to the determination of the face identity. CONSTITUTION: A filter designing device determines one or more face population groups and selects one or more designing parameters with respect to a face recognition filter (S100,S110). A device separately applies a virtual filter corresponding to a relevant design combination to images contained in a face population group according to the combination of the designing parameters (S130,S140). The device calculates intra-individual variances, inter-individual variances, and objective values by the designing combinations (S150,S160). The device determines a face recognition filter to design (S180). [Reference numerals] (1) Filter designing device; (AA) No; (BB) Yes; (S100) Determine one or more face population groups which contains images obtained from different shooting conditions; (S110) Select a designing parameter; (S120) Determine the application range and step size with respect to the designing parameter; (S130) Form a plurality of designing combinations; (S140) Apply a virtual filter to a face population group according to a single designing combination; (S150) Calculate intra-individual variances and inter-individual variances on the basis of filter responding data; (S160) Calculate objective values on the basis of the individual variances and the inter-individual variances; (S170) Tasks carried out with respect to all designing combinations?; (S180) Select a maximum value among a plurality of objective values and select a designing combination corresponding to the value

Description

Face identification filter planning method and apparatus

The present invention relates to a method and a device for designing a filter, and more particularly, to a method and device for designing a filter for recognizing a face.

Face recognition among biometrics is applied to various fields, and various studies are being conducted to increase the accuracy of face recognition. Filtering to enable quantization in the frequency domain is widely used for computerized identification of acquired face images. The face recognition filter used for filtering includes a Gabor filter.

In general, a filter bank including a plurality of filters having a series of frequencies and directional factors as a factor is used rather than a single filter. Directional values of each filter in the filter set may be arbitrarily selected. Since the frequency is a continuous quantity, the selectable frequencies for the filter set are infinite. When designing the face recognition filter in consideration of this infinite design space, it is necessary to verify the suitability and achievement of the optimization criterion assumed by the factor change and the designer's change according to the factor change. This verification is an area that needs to be monitored interactively. Considering these aspects, there is no environment for developing a face recognition filter.

The filter set may exhibit various sensitivity to various image characteristics. For example, the image quality of the image may show various changes (eg, low quality or high quality) according to the image capturing apparatus and shooting conditions such as lighting degree, resolution, and distance. Since these various changes according to the shooting conditions are not related to the face identity of the individual, the ideal face recognition filter is required to have a minimum sensitivity to the changes according to the shooting conditions.

On the other hand, a population (set), which is a set of face images for multiple people, has different variations (variations, variances) by face part (e.g. eyes, nose, mouth) and by the frequencies of the filter set used. . An ideal face recognition filter is required to maintain maximum sensitivity to this dispersion.

Given that the design space for the factor is infinite, interactive monitoring of the change of these objective functions according to the change of the factor is required for efficient filter design.

The technical problem to be solved by the present invention is to provide a method and apparatus for easily designing an ideal filter.

Another object of the present invention is to provide a method and an apparatus capable of designing an optimal filter while efficiently changing factors for the filter through interaction with a designer.

A method according to an aspect of the present invention is a method of designing a face recognition filter, comprising: determining at least one face population set including images of a plurality of people to which the face recognition filter to be designed is to be applied; Selecting at least one design parameter for the face recognition filter and combining the selected design parameters to generate at least one design combination; For each of the design combinations, applying a virtual filter corresponding to the design combination to each of the images included in the face population set to obtain filter response data; For each design combination, calculating intra-individual variance values representing differences of each image for the same person and inter-individual dispersion values representing image differences between different people based on the obtained filter response data; Calculating an objective function value for each design combination based on the in-person dispersion value and the in-person dispersion value; And determining a face recognition filter to be designed based on the objective function values for each design combination.

The determining of the at least one face population set may determine a plurality of face population sets including images of a plurality of people obtained under different photographing conditions. Here, the plurality of face population sets may each include images acquired under different shooting conditions for the same person.

In this case, the determining of the at least one face population set may include: outputting an image selection input window for selecting a shooting condition; Selecting a shooting condition based on the data input through the image selection input window, wherein the shooting condition is at least one of an illumination degree, a resolution, a distance, and a camera angle; And determining a face population set corresponding to the selected photographing condition.

The generating of the design combination may include selecting a design parameter which is at least one of a frequency and directivity of a face recognition filter and a width and a size of a kernel corresponding to an area of an image to which the filter is applied; And setting a plurality of values for the selected design parameters, and generating a plurality of design combinations by combining a plurality of values of the set design parameters.

Wherein the generating of the plurality of design combinations comprises: determining an application range and step size consisting of a minimum value and a maximum value for each of the selected design parameters; Extracting a plurality of values corresponding to the values corresponding to the step size within an application range for each design parameter; And generating a plurality of design combinations by combining the plurality of values extracted for each design parameter.

The generating of the design combination may include: outputting a design parameter input window for selecting an application range and a step unit for each design parameter; And selecting at least one design parameter based on data input through the design parameter input window, and determining an application range and a step unit for each selected design parameter.

The calculating of the variance values may include calculating an variance value between individuals for the people included in the face population set based on filter response data corresponding to the images of the plurality of people included in the face population set. Calculating interpersonal variance values for each of all face population sets; And calculating an in-person variance value for one person based on filter response data corresponding to images corresponding to one person from the plurality of face population sets, and calculating an in-person variance value for each person. It may include the step.

In this case, the calculating of the objective function value may calculate the objective function value for each design combination by dividing the variance value between the individuals by the variance value within the individual.

The calculating of the objective function may include calculating an objective function based on objective function = max [sum M (Var_bj) / sum N (Var_wi)], where sum M (Var_bj) is a set of all face populations. The sum N (Var_wi) represents the sum of the intra-person variance values Var_bj calculated for each person, and the sum N (Var_wi) represents the sum of the intra-person variance values Var_wi calculated for all people.

The determining of the face recognition filter may include selecting an objective function value having a maximum value among the objective function values acquired for each design combination, and designing a face recognition filter based on a design combination corresponding to the selected objective function value. have.

In addition, the filter design apparatus according to another aspect of the present invention, the image set determination unit for determining at least one face population set including images of a plurality of people to apply the face recognition filter to be designed; A parameter selector to select at least one design parameter for the face recognition filter; A design combination generation unit generating at least one design combination by combining the selected design parameters; A virtual filter applying unit for obtaining filter response data for each of the design combinations by applying a virtual filter corresponding to the design combination to the images included in the face population set; A filter feature variable calculating unit for calculating individual variance values representing differences between individual images of the same person and individual variance values representing image differences between different people based on the acquired filter response data; And an objective function calculator configured to calculate an objective function value for each design combination based on the in-person variance value and the interpersonal variance value, and determine a face recognition filter to be designed based on the objective function value for each design combination.

In addition, the apparatus may further include an image storage unit configured to store a plurality of face population sets including images of a plurality of people acquired under different shooting conditions, wherein the plurality of face population sets may be different from each other in the same person. Each of the images obtained from the may include.

The apparatus may further include a parameter range determination unit configured to determine an application range and a step size including a minimum value and a maximum value for each of the selected design parameters. In this case, the design combination generation unit may include the step size within the application range for each design parameter. A plurality of design combinations may be generated by extracting a plurality of values corresponding to the values corresponding to and combining the plurality of values for each design parameter.

The parameter selector may select a design parameter that is at least one of a frequency and direction of the face recognition filter, and a width and a size of a kernel corresponding to an area of an image to which the filter is applied.

The objective function calculator may select an objective function value having a maximum value among the objective function values obtained for each design combination, and design a face recognition filter based on the design combination corresponding to the selected objective function value.

In addition, the filter design apparatus includes an image selection input window for selecting shooting conditions, a design parameter input window for selecting an application range and a step unit for each design parameter, and an output unit for outputting the filter response data and the objective function value; And an input unit configured to transfer data input through the image selection input window and the design parameter input window to the image set determiner, the design parameter selector, and the parameter range determiner, respectively.

According to an embodiment of the present invention, an optimal face recognition filter design is achieved.

In addition, even in an environment where the factors of the face recognition filter can be infinitely designed, the sensitivity of the face recognition filter has the maximum sensitivity to the variance representing the face identity of different people and minimum sensitivity to the matters not related to the face identity determination. Face recognition filters can be designed. Therefore, a filter having improved face recognition characteristics can be easily designed.

1 is a flowchart illustrating a filter design method according to an embodiment of the present invention.
2 is an exemplary view showing filter response data according to an exemplary embodiment of the present invention.
3 to 5 are exemplary views showing the degree of dispersion in accordance with the characteristic amount of the filter according to an embodiment of the present invention.
6 is a structural diagram of a filter design device according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention.

In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and like reference numerals designate like parts throughout the specification.

Throughout the specification and claims, when a section is referred to as including an element, it is understood that it may include other elements as well, without departing from the other elements unless specifically stated otherwise.

A method and apparatus for designing a face recognition filter according to an exemplary embodiment of the present invention will now be described.

The face recognition filter includes a plurality of factors, and the factors include frequency, orientation, and the like. Since the characteristics of the face recognition filter change according to the factors, the face recognition filter having the desired characteristics may be obtained only by optimally selecting the factors. In other words, it is necessary to determine the frequency and direction of the filter appropriately to obtain a face recognition filter having desired characteristics.

The face recognition filter has various response values, ie sensitivity, for various image characteristics. The image may have various characteristics according to shooting conditions (eg, illumination, resolution, distance, etc.), and may have a change in image quality such as low quality, neutral quality, and high quality according to the shooting conditions. Since the image characteristics such as the image quality change according to the shooting conditions are not substantially related to the identity that is different for each individual, the smaller the sensitivity of the face recognition filter for such image characteristics, the more accurate face recognition can be achieved.

On the other hand, when the face recognition filter is applied to the image, the variation in the face population based on the pixel values obtained after filtering by the face region (for example, eyes, nose, mouth, etc.) of the image is an individual's identity. It is used for personal identification as it relates to the identification of. Therefore, the greater the sensitivity of the face recognition filter to such dispersion, the more accurate face recognition can be achieved.

The face recognition filter should be designed to have a uniform response value, that is, sensitivity, to the image of the same person, and to be as sensitive as possible to the image of others. An embodiment of the present invention provides a method and apparatus for designing a face recognition filter such that an optimal condition for the face recognition filter is satisfied.

For face recognition, a filter set (or filter bank) including a plurality of face recognition filters (hereinafter referred to as "filters" for convenience of description) is used. The filter set may be characterized by combining factors for each of the filters included, for example, a filter set comprising 24 filters may be used if it has eight directional, three frequencies. .

For filter design, a face population set is used that includes images of faces with statistical representation. The face population set may be classified according to image characteristics. For example, the face population set may be classified according to the image quality of the image. As a specific example, it may be classified into a face population set made of images having low quality, a face population set made of images having neutral quality, and a face population set made of images having high quality. In this case, each image included in the face population set for each image quality may be an image of the same person. That is, the first image of the face population set corresponding to the low quality, the first image of the face population set corresponding to the medium quality, and the first image of the face population set corresponding to the high quality may be images of the same person.

In an exemplary embodiment of the present invention, face population sets including images acquired according to different shooting conditions (eg, lighting degree, resolution, distance, etc.) of various people are used. For example, face population sets s1, s2, s3, ..., sm obtained for n people (1, 2, 3, ..., n) according to various shooting conditions (c1, c2, c3, ..., cm). Are used. An identifier may be assigned to each face population set, and an identifier may also be assigned to each image included in the face population set. For example, the face population set having an identifier of s1 is id001, id002, id003,... , may include images having an identifier such as idn. Particularly, in the embodiment of the present invention, for each image included in the face population set for each shooting condition, the identifier may be identified under which shooting condition the image is obtained, for example, "id001_lighting010_resolution 030. You can also use an identifier of the form "." That is, the identifier of the image included in the face population set may be implemented to include feature data according to shooting conditions.

In an embodiment of the present invention, the filter may be designed through a combination of design parameters. The design parameters here include a factor and kernel that determine the characteristics of the filter. The kernel specifies the height and width of a given area of the image to which the filter is applied.

1 is a flowchart illustrating a face recognition filter design method according to an exemplary embodiment of the present invention. Such a filter design method may be performed by the filter design device 1.

First, as shown in FIG. 1, the filter design apparatus 1 determines a face population set that is images to which the face recognition filter to be designed is applied. In particular, the face population set including the images having various values in the characteristics of the image that the filter to be designed has the minimum sensitivity is determined (S100). For example, in order to design a filter with minimal sensitivity to image quality, a face population set including images acquired under low quality shooting conditions for each of n people, and a face population including images obtained under medium quality shooting conditions The face population sets including the set and the images obtained under the high quality shooting conditions are determined. Here, the low quality shooting conditions, the medium quality shooting conditions, and the high quality shooting conditions have preset illumination degree, resolution, distance, or camera angle for each condition. In addition, while selecting values such as illumination, resolution, distance, or camera angle individually as a separate value instead of a preset value, a desired shooting condition may be selected.

Here, the filter design device 1 may provide an image selection input window through output means so that a designer can select images having various values corresponding to desired image characteristics. The image selection input window is configured to individually select or input lighting conditions, resolution, distance, or camera angle, which are photographing conditions. That is, for each illumination degree, resolution, distance, or camera angle constituting the shooting condition, an image selection input window may be provided to select one of a plurality of preset values or to input an arbitrary value. . The filter design device 1 stores from the storage means a face population set including images acquired under corresponding shooting conditions based on the values of illumination degree, resolution, distance, or camera angle input through the image selection input window. Load.

As described above, after determining a plurality of face population sets including images having various values in the characteristics of the image to which the filter to be designed has the minimum sensitivity, design parameters are determined (S110).

The design parameters include the frequency and directivity of the filter, and also include the width and size of the kernel, and may use at least one of these design parameters. After the design parameters to be used for the filter design are determined, an application range and step size for the corresponding design parameters are determined (S120). Here, the application range is a range having the maximum value and the minimum value for the design parameter, and the step unit represents the interval at which the corresponding values change in the application range. For example, when the frequency of the filter and the width of the kernel are used as design parameters, "0.1" may be determined in units of steps while an application range of 0.1 to 0.3 (cycles / pixel) is set for the frequency. Accordingly, values of "0.1, 0.2, 0.3" are used as the frequency. In addition, "2" may be determined in units of steps while an application range of 11 to 19 is set for the kernel width. Accordingly, values of "11, 13, 15, 17, 19" are used as the width of the kernel.

Here, the filter design device 1 may provide a design parameter input window through an output means so that a designer can select a desired design parameter and an application range and a step unit for each parameter. In this case, a design parameter input window may be provided to select one of a plurality of preset values for the application range and the step unit for each design parameter or to input an arbitrary value.

As described above, after at least one design parameter is determined, an application range and a step unit are determined for each design parameter to obtain a plurality of values for each design parameter. A plurality of design combinations may be obtained by combining the design parameters having the plurality of values (S130). For example, combining the three frequency values and the values of five kernel widths as described above, 15 design combinations can be obtained. Virtual filters corresponding to these design combinations, that is, virtual filters designed with design parameter values according to the design combinations, form a filter set.

The process of selection and determination as described above can be performed through an interactive interface with the designer designing the filter.

The filter design apparatus 1 obtains the response of the filter corresponding to the design combination by applying the acquired design combinations to the face population sets which are the image sets determined in the above step S100 (S140). Here, the filter response is referred to as "filter response data" for convenience of explanation. Virtual filters having different characteristics for each design combination are applied to each image included in the face population set, thereby obtaining filter response data resulting from applying the virtual filter to each image for each face population set. To this end, the filter design device 1 performs sweeping for all design combinations formed on the basis of design data. For example, all hypothetical filters corresponding to 15 design combinations obtained by combining three frequency values and five kernel width values can be obtained from a low quality face population set, a medium quality face population set, and a high quality face set. Apply for each image of, respectively, to obtain filter response data for each image. 2 is an exemplary view showing filter response data according to an exemplary embodiment of the present invention. Here, filter response data obtained when a virtual filter for each design combination composed of Gabor filters is applied to an image is shown.

Next, the filter design apparatus 1 calculates variance values representing differences for each image of the same person, based on filter response data obtained by applying virtual filters according to a design combination to each face population set, and also mutually. Calculate variances that represent image differences between different people. The variance value representing the difference for each image of the same person may be referred to as an "intra-individual variance value", and the variance values representing the image difference between different people may be referred to as the "dispersion value between individuals" (S150).

For example, filter response data obtained by applying a design parameter to a face population set may be shown in Table 1 below. Table 1 below shows a case where a virtual filter according to one design combination is applied to a plurality of face population sets.

For example, suppose there are 15 design combinations obtained by combining three frequency values and five kernel width values. As described above, the personal identifiers 001, 002, 003,... A plurality of face population sets (s1, s2, s3, ..., sm) containing images obtained under different shooting conditions for n persons (1, 2, 3, ..., n) each having n Let's say When the virtual filter according to the design combination is applied to each image of the plurality of face population sets, filter response data as shown in Table 1 described above may be obtained.

If the virtual filter according to the first design combination is applied to each image of the plurality of face population sets s1, s2, s3, ..., sm, among the 15 design combinations, the filter response data R1_001_S1 and R1_001_S2 are shown in Table 1. ... R1_001_Sm, R2_002_S1, R2_002_S2, ... R2_002_Sm, ..., Rn_n_S1, Rn_n_S2, ... Rn_n_Sm) are obtained.

The filter design device 1 calculates the in-person dispersion value and the in-person dispersion value based on the filter response data shown in Table 1 obtained by applying the virtual filter according to the first design combination to the face population sets.

That is, as a result of applying the virtual filter according to the first design combination to the plurality of images included in the face population set according to different shooting conditions for the individual having the identifier of 001, R1_001_S1, R1_001_S2,... Filter response data of R1_001_Sm are obtained. Since these filter response data correspond to one person having an identifier of 001, the filter design device 1 calculates an in-person variance value Var_w1 for a person having an identifier of 001 based on such filter response data.

As a result of applying the virtual filter according to the first design combination to the plurality of images included in the face population set according to different shooting conditions for the individual having the identifier of n, Rn_n_S1, Rn_n_S2,... Filter response data of Rn_n_Sm is obtained, and an intraperson dispersion value Var_w1 for an individual having an identifier of n is calculated based on these filter response data. That is, variance values are calculated for each individual.

Meanwhile, according to Table 1 above, for each of the images included in the face population set having the identifier of s1, R1_001_S1, R2_002_S1,... Filter response data such as Rn_n_S1 is obtained. The filter design apparatus calculates the variance value Var_b1 between different individuals based on the filter response data obtained for the images included in the face population set of s1. Thus, the variance value Var_bm between individuals for the face population set of Sm is calculated based on the filter response data obtained for each image included in the face population set having the identifier of sm. In other words, the variance value between individuals is calculated for each face population set.

3 to 5 are exemplary views showing the degree of dispersion in accordance with the characteristic amount of the filter according to an embodiment of the present invention.

3 to 5, each graph applies a virtual filter (eg, a Gabor filter, a local binary pattern (LBP), a uniform LBP, etc.) according to a design combination to each face in the face population set. The Gabor filters have angle and frequency characteristics, and FIGS. 3 to 6 show the filter images according to the design combinations having the same frequency but different angles. Represent variances of filter response data, this is just one example.

Next, the filter design apparatus 1 calculates an objective function corresponding to the corresponding design combination based on the interpersonal variance values and the in-person variance values acquired based on the virtual filter according to one design combination (S160). The objective function can be written as

Here, sum M (Var_bj) represents the sum (first total variance) of the interpersonal variance values Var_bj calculated for all face population sets, and sum N (Var_wi) is respectively calculated for all individuals. The sum of the intra-person variances Var_wi (the second total variance).

Therefore, a value calculated by dividing the first total variance value by the second total variance value is stored as the objective function value. At this time, the filter design device 1 displays the calculated objective function value through the output means, so that the designer is a feature amount of the filter according to the variance representing the difference between the individuals according to the design combination and the variance representing the difference within the individual. Allows you to check the objective function value.

In this way, the process of obtaining the filter response data and calculating the intra-individual variance value, the inter-individual variance value, and the objective function value is performed on all the design combinations, for example, each of the 15 design combinations (S170).

Here, the filter design device 1 may calculate the in-person dispersion value and the in-person dispersion value using only a part of all the filter response data obtained. For example, for a face population set containing 64 × 64 face images, applying a set of filters combining eight directional values and three frequency values results in approximately 100,000 (98304 = 64 × 64 × 8). Filter response data of × 3) may be obtained. To improve the interpretability and ease of processing in processing such a large amount of filter response data, some filter responses, for example, satisfying a set condition or randomly extracting a predetermined condition instead of using all filter response data, are used. The data can be used to calculate the variance in individuals, the variance between individuals, and the objective function.

Meanwhile, when the process of calculating the objective function values for all the design combinations is completed, the filter design apparatus 1 selects the objective function value having the largest value among the objective function values calculated for each design combination. The filter is then designed according to the selected combination of the calculated objective function values.

On the other hand, if the objective function value for one design combination is calculated while calculating the objective function value for each design combination, the current calculated value is compared with the previously stored objective function value and the objective function value is transferred if it is smaller. If the current calculated value is larger than the previously stored objective function value, the objective function value for all design combinations is calculated by storing the current calculated value as the objective function value. Upon completion, the objective function value having the largest value may be finally stored.

These objective function values allow the design of a face recognition filter with maximum sensitivity for variances representing face identity for different people, and minimum sensitivity for matters not related to facial identity determination.

Next, the structure of the filter design apparatus is demonstrated.

6 is a view showing the structure of a filter design device according to an embodiment of the present invention.

As shown in FIG. 6, the filter design apparatus 1 according to an exemplary embodiment of the present invention may include an image set determiner 11, a design parameter selector 12, a parameter range determiner 13, and a design combination generator. 14), a virtual filter applying unit 15, a filter feature variable calculating unit 16, an objective function calculating unit 17, an image storing unit 18, and an input unit 19 and an output unit 20. The image storage unit 18 stores face population sets including images acquired for a plurality of people according to various shooting conditions. The input unit 19 functions as an input interface that receives input from the designer and transfers it to each of the units 11, 12, and 13, and the output unit 20 allows the designer to check the processing results of each unit 11 to 17. It acts as an output interface to display or output.

The image set determiner 11 determines a face population set, which is an image to which the face recognition filter to be designed is applied, and includes images having various values in the characteristics of the image that the filter to be designed has with minimum sensitivity. Determine face population sets. The determined facial population sets may be loaded from the image storage unit 18 and provided to the virtual filter applying unit 15. Here, the image set determiner 11 may provide an image selection input window through the output unit 20 so that a designer can select images having various values corresponding to desired image characteristics.

The design parameter selector 12 selects a design parameter that is at least one of the frequency and directivity of the filter, or the width and size of the kernel. The parameter range determination unit 13 determines the application range and the step size for the design parameter. Here, the parameter range determination unit 13 may provide a design parameter input window through the output unit 20 so that a designer can select a desired design parameter and an application range and step unit for each parameter.

The design combination generation unit 14 generates a plurality of design combinations by combining a plurality of values for each design parameter obtained based on an application range and a step unit determined for each design parameter.

The virtual filter application unit 15 applies virtual filters corresponding to the generated design combinations to each of the images included in the face population set, and is a filter that is a result of applying the virtual filter to each image for each face population set. Acquire response data.

The filter feature variable calculating unit 16 is based on the filter response data obtained by applying the virtual filters according to the design combinations to each face population set. Values, variance values between individuals representing the image differences between individuals are calculated.

The objective function calculation unit 17 calculates the objective function value by dividing the second total variance value, which is the sum of the variance values in the individual, by the first total variance value, which is the sum of the variance values between individuals. The maximum value is selected from the calculated objective function values. In addition, the calculated objective function values and the objective function value of the selected maximum value may be displayed through the output unit 20.

Here, the objective function value having the maximum value may be selected from among the objective function values acquired for each design combination, and the face recognition filter may be designed based on the design combination corresponding to the selected objective function value.

Although the present invention has been described and illustrated in detail, it is clearly understood that the same is by way of illustration and example only and is not to be construed as limiting the scope of the present invention. Readable recording medium. The present invention can be easily implemented by those skilled in the art from the description of the embodiments described above.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, It belongs to the scope of right.

Claims (17)

In how to design face recognition filter,
Determining at least one face population set including images of a plurality of people to which the face recognition filter to be designed is to be applied;
Selecting at least one design parameter for the face recognition filter and combining the selected design parameters to generate at least one design combination;
For each of the design combinations, applying a virtual filter corresponding to the design combination to each of the images included in the face population set to obtain filter response data;
For each design combination, calculating intra-individual variance values representing differences of each image for the same person and inter-individual dispersion values representing image differences between different people based on the obtained filter response data;
Calculating an objective function value for each design combination based on the in-person dispersion value and the in-person dispersion value; And
Determining a face recognition filter to be designed based on the objective function values for each design combination;
≪ / RTI > The method of claim 1, wherein
Determining at least one face population set,
Determine a plurality of face population sets including images of a plurality of people acquired under different shooting conditions, wherein the plurality of face population sets each include images acquired under different shooting conditions for the same person; . The method according to claim 2, wherein
Determining at least one face population set,
Outputting an image selection input window for selecting shooting conditions;
Selecting a shooting condition based on the data input through the image selection input window, wherein the shooting condition is at least one of an illumination degree, a resolution, a distance, and a camera angle; And
Determining a face population set corresponding to the selected shooting condition
≪ / RTI > The method of claim 1, wherein
Generating the design combination,
Selecting a design parameter that is at least one of a frequency and a direction of the face recognition filter and a width and a size of a kernel corresponding to an area of an image to which the filter is applied; And
Setting a plurality of values for the selected design parameters and generating a plurality of design combinations by combining the plurality of values of the set design parameters
≪ / RTI > The method of claim 4, wherein
Generating the plurality of design combinations
Determining an application range and step size consisting of minimum and maximum values for each of the selected design parameters;
Extracting a plurality of values corresponding to the values corresponding to the step size within an application range for each of the design parameters; And
Generating a plurality of design combinations by combining a plurality of values extracted for each design parameter
≪ / RTI > The method of claim 4, wherein
Generating the design combination,
Outputting a design parameter input window for selecting an application range and a step unit for each design parameter; And
Selecting at least one design parameter based on data input through the design parameter input window, and determining an application range and a step unit for each selected design parameter
≪ / RTI > The method of claim 1, wherein
The calculating of the variance values
Based on the filter response data corresponding to the images of the plurality of people included in one face population set, the variance value of individuals for the people included in the face population set is calculated, and the variance between individuals for each of the face population sets is calculated. Calculating values; And
Computing an in-person variance for one person based on filter response data corresponding to images corresponding to a person from the plurality of face population sets, calculating an in-person variance for each individual step
≪ / RTI > The method of claim 7, wherein
Computing the objective function value
And calculating an objective function value for each design combination by dividing the variance value between the individuals by the variance value within the individual. The method of claim 8, wherein
Computing the objective function value
Objective Function = max [sum M (Var_bj) / sum N (Var_wi)]
Is based on
The sum M (Var_bj) represents the sum of the interpersonal variance values Var_bj calculated for all face population sets, and the sum N (Var_wi) represents the intra-person variance values Var_wi calculated for all people, respectively. Indicating the sum of. The method of claim 1, wherein
Determining the face recognition filter
Selecting an objective function value having a maximum value among the objective function values obtained for each design combination, and designing a face recognition filter based on a design combination corresponding to the selected objective function value. An image set determiner configured to determine at least one face population set including images of a plurality of people to which a face recognition filter to be designed is applied;
A parameter selector to select at least one design parameter for the face recognition filter;
A design combination generation unit generating at least one design combination by combining the selected design parameters;
A virtual filter applying unit for obtaining filter response data for each of the design combinations by applying a virtual filter corresponding to the design combination to the images included in the face population set;
A filter feature variable calculating unit for calculating individual variance values representing differences between individual images of the same person and individual variance values representing image differences between different people based on the acquired filter response data; And
An objective function calculation unit configured to calculate an objective function value for each design combination based on the in-person variance value and the interpersonal variance value, and determine a face recognition filter to be designed based on the objective function value for each design combination.
Including, filter design device. The method of claim 11, wherein
The apparatus may further include an image storage unit configured to store a plurality of face population sets including images of a plurality of people acquired under different shooting conditions.
And the plurality of face population sets each include images acquired under different photographing conditions for the same person. The method of claim 11, wherein
And a parameter range determiner configured to determine an application range and step size consisting of minimum and maximum values for each of the selected design parameters.
The design combination generating unit extracts a plurality of values corresponding to the values corresponding to the step size within an application range for each design parameter, and generates a plurality of design combinations by combining the plurality of values for each design parameter. Device. The method of claim 11, wherein
And the parameter selector selects a design parameter that is at least one of a frequency and a direction of a face recognition filter and a width and a size of a kernel corresponding to an area of an image to which the filter is applied. The method of claim 11, wherein
The objective function calculation unit
Objective Function = max sum M (Var_bj) / sum N (Var_wi)
To calculate the value of the objective function,
The sum M (Var_bj) represents the sum of the interpersonal variance values Var_bj calculated for all face population sets, and the sum N (Var_wi) represents the intra-person variance values Var_wi calculated for all people, respectively. Filter design device, representing the sum of. The method of claim 11, wherein
The objective function calculation unit
And selecting an objective function value having a maximum value among the objective function values obtained for each design combination, and designing a face recognition filter based on a design combination corresponding to the selected objective function value. The method of claim 13, wherein
An image selection input window for selecting shooting conditions, a design parameter input window for selecting an application range and a step unit for each design parameter, and an output unit for outputting the filter response data and the objective function value; And
An input unit for transferring data input through the image selection input window and the design parameter input window to the image set determiner, the design parameter selector, and the parameter range determiner, respectively.
Further comprising a filter design device.

Images