KR101471204B1 - Apparatus and method for detecting clothes in image - Google Patents

Abstract

The present invention discloses a garment detecting apparatus for recognizing a face of a person in an image and detecting clothing based on the recognized face and a method thereof. The present invention relates to a face recognition apparatus for recognizing a face of a person in a query input image input as a query. A garment region extracting unit for extracting a garment region from a query input image based on the recognized face; A color similarity calculation unit for color matching the extracted garment area to a predetermined area of the comparison object image and calculating a color similarity value of the garment area from the color matching; A texture similarity calculation unit for matching the extracted garment area with a predetermined area of the comparison target image and calculating a texture similarity value of the garment area from the texture matching; And comparing the calculated color similarity value and the calculated texture similarity value with a predetermined reference value to determine whether the color or texture of the extracted garment area is similar to a predetermined area of the comparison object image. If the discrimination result is similar, And a garment detecting unit that detects the garment detecting unit as a garment image representing the garment. According to the present invention, it is possible to improve the recognition rate of clothes worn by a person in an image. In addition, various service competitiveness based on video objects can be secured based on metadata.

 Face recognition, clothing detection, metadata, color matching, texture matching, histogram intersection,

Description

[0001] APPARATUS AND METHOD FOR DETECTING APPAREL IN IMAGE [0002]

The present invention relates to an apparatus and method for detecting garments in an image. More particularly, the present invention relates to a clothes detecting apparatus for recognizing a face of a person in an image and detecting clothing based on the recognized face, and a method thereof.

In the case of purchasing conventional clothes, the user first visits a department store, a distributor, an apparel shop, and selects a specific garment based on taste, preference, and personality. After that, the apparel was purchased only after wearing the selected clothes and confirming the purchase intention. Recently, as the Internet has been activated, websites such as Internet shopping malls have been increasing, and devices capable of accessing the Internet from time to time, such as PCs and mobile phones, have become popular.

However, in the conventional Internet shopping mall, in order to register the clothes on the website, it was necessary to photograph the clothes directly or to obtain photographs of the clothes. However, it was impossible to have various kinds of clothes, and it was not easy to obtain photographs taken at various angles. In the past, image processing technology was used to recognize garments in images.

However, among conventional image processing techniques, a method of recognizing clothes in a video has the following problems. First, because it uses the global inspection method for HD-class video or Full-HD video, it takes much time to process the video. Second, because of the influence of lighting and the angle of camera shooting, the clothing itself causes many changes.

Meanwhile, with the growth of the video content market today, the video service market through various media is gradually being activated. In response to this trend, in recent years, a video for a new interactive service and various players (e.g., a digital broadcast player) for executing the video have been developed in advance. Also, infrastructures are gradually being built so that these services can be implemented.

However, in order to implement such an interactive service, a procedure for generating metadata for interactivity appropriate to the contents of moving picture data is required. In addition, there is a need for an authoring tool for creating such metadata. However, the existing metadata authoring tools allow the metadata editor to directly set the object data in the video, and to generate the metadata from the metadata. Therefore, in the past, it was inconvenient for the editor to watch the whole contents of the video.

Recently, as the UCC service has become generalized, a Web service has emerged in which metadata is generated through tracking after selecting objects in the UCC video. The sites that provide these services include asterpix.com, overlay.tv, and videoclix.tv. Recently, such video service is considered to be applied to IPTV and the like. However, the conventional metadata authoring tool has the following problems. First, it is very difficult for beginners to use because it is necessary to understand the contents in the video and select the object to edit directly. Second, in order to edit HD / Full-HD video for IPTV service, editing time is 5 ~ 10 times compared with video playback time.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a garment detecting apparatus for detecting a face of a person in an image and detecting the garment based on the recognized face, and a method thereof.

It is another object of the present invention to provide a garment detecting apparatus and method for automatically generating garment-related metadata using the garment detecting method.

According to an aspect of the present invention, there is provided a face recognition apparatus comprising: a face recognition unit for recognizing a face of a person in a query input image input as a query; A garment region extracting unit for extracting a garment region from the query input image based on the recognized face; A color similarity calculation unit for color matching the extracted garment area to a predetermined area of a comparison object image and calculating a color similarity value of the garment area from the color matching; A texture similarity calculation unit for matching the extracted garment area with a predetermined area of a comparison target image and calculating a texture similarity value of the garment area from the texture matching; And comparing the calculated color similarity value and the calculated texture similarity value with a predetermined reference value to determine whether the color or texture of the extracted garment area is similar to a predetermined area of the comparison object image, And a garment detecting unit for detecting the garment area as a garment image representing the garment, if the garment detecting unit detects the garment area.

Preferably, the garment region extractor sets the face region based on the size of the face, and extracts the garment region from a position below the set face region.

More preferably, the garment region extracting unit may include a face region size determining unit for determining a horizontal size and a vertical size of the set face region; And a garment region detecting unit for detecting the garment area having a width of 2.5 to 3.5 times the width of the face area at a position below the face area and a height of 1.5 to 2.5 times the vertical size of the face area .

Preferably, the color similarity calculation unit includes: a color space conversion unit for converting the color space of the frame including the garment area from RGB to HSI; A color information deriving unit for deriving information on a color in consideration of the RGB in the converted image; And a color similarity value calculating unit for comparing the two color information and calculating the color similarity value, when information on the color is derived from a predetermined area of the comparison object image.

More preferably, the color similarity value calculation unit divides the query input image into blocks, extracts at least one block from the divided blocks, and compares the extracted block with a block of the comparison object And the two color information are compared through block matching. The color information derivation unit may derive a color value in the garment area including a R (Red) component value, a G (Green) component value, and a B (Blue) component value as information on the color, Normalize the color value to a range value from 0 to 12.

Preferably, the texture similarity calculator includes: a frequency domain switching unit for converting an image of the garment area into a frequency domain using DCT; A texture feature vector extractor for extracting a frequency domain coefficient from among transform coefficients for each block of the image derived from the conversion and extracting a texture feature vector including the extracted frequency domain coefficients; A distance calculator for comparing the extracted at least two texture feature vectors to obtain a distance between texture feature vectors; And a texture similarity value calculator for calculating the texture similarity value by comparing the distance values obtained when the distance between the texture feature vectors is determined from a predetermined area of the comparison object image.

More preferably, the distance calculator calculates the distance between the texture feature vectors using nine texture feature vector values that exist for each block.

Preferably, the clothes detecting device further includes a person information acquiring unit that acquires information about a person indicated by the recognized face.

According to another aspect of the present invention, there is provided an image processing method comprising the steps of: (a) recognizing a face of a person in a query input image input as a query; (b) extracting a garment area from the query input image based on the recognized face; (c) color matching the extracted garment area to a predetermined area of a comparison image, and calculating a color similarity value of the garment area from the color matching; (d) texture matching the extracted garment area to a predetermined area of a comparison object image, and calculating a texture similarity value of the garment area from the texture matching; And comparing the calculated color similarity value and the calculated texture similarity value with a predetermined reference value to determine whether the color or texture of the extracted garment area is similar to a predetermined area of the comparison object image, And detecting the garment area as a garment image representing the garment if the results are similar.

Preferably, the intermediate step between the step (d) and the step (e) includes the step of acquiring information about a person indicated by the recognized face, and the step (e) And uses information on the acquired person.

According to the present invention, the following effects can be obtained. First, recognizing a person's face in an image and detecting the garment based on the recognized face can improve the recognition rate of the clothing worn by the person. In particular, it is possible to develop a garment recognition technology having an excellent recognition rate of clothes and a fast execution speed by searching clothes only for the frames whose faces are recognized without matching garments in all video frames. Second, garment-related metadata is automatically generated by using the garment detection method, so that the metadata-based metadata authoring time can be greatly shortened. Also, by shortening metadata authoring time, various service competitiveness based on video objects can be ensured.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that, in adding reference numerals to the constituent elements of the respective drawings, the same constituent elements are denoted by the same reference symbols as possible even if they are shown in different drawings. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. In addition, the preferred embodiments of the present invention will be described below, but it is needless to say that the technical idea of the present invention is not limited thereto and can be variously modified by those skilled in the art.

1 is a block diagram schematically showing a clothes detecting apparatus according to a preferred embodiment of the present invention. 2 is a flowchart illustrating a method of detecting a garment according to a preferred embodiment of the present invention. 1, a clothing detection apparatus 100 according to a preferred embodiment of the present invention includes a face recognition unit 110, a garment region extraction unit 120, a color similarity calculation unit 130, A person detection unit 160, a power source unit 170, and a control unit 180. The control unit 180 controls the operation of the display unit 140, the person information acquisition unit 150,

Recently, as interactive services such as IPTV are being implemented, there is a need for technology for image-based interactive services. For these services, automatic recognition of objects is the key. The clothes detecting apparatus 100 according to the present invention performs a function of automatically recognizing clothes among various objects of an image. The use of the clothes detecting apparatus 100 can shorten the authoring time of the authoring tool for an interactive service based on a video object such as generation of garment related meta data. If the garment object recognition based on the face recognition is automatically performed by overcoming the limitation of the method which has previously specified the object area through the existing manual operation, the extraction time of the moving object object metadata, that is, the moving object metadata, It is possible.

The garment recognition technology implemented by the garment detecting device 100 assumes that only clothes that are worn by a person are meaningful because the garment recognition technology does not maintain a uniform shape even if the garment has the same shape.

The face recognition unit 110 performs a function of automatically recognizing a face of a person in an image such as a moving image by using a tool for automatically recognizing the face. When the face recognizing unit 110 automatically recognizes the face of the person, the apparel area extracting unit 120 automatically extracts the apparel area located at the bottom of the face. When the apparel area extractor 120 automatically searches for the apparel area, it is based on the size of the face and the ratio of the upper body of the person.

When the position of the face is determined in the image and the garment area is extracted from the image, the garment can be found for the entire area of the moving image based on the extracted garment area. Color information and texture information, which are characteristics of clothes, are used when searching for clothes. If clothes are found using only the color information of the clothes, there is a high possibility that an error such as being erroneously detected due to the influence of illumination or the like occurs. In addition, if clothing is searched using only the texture information of the clothes, an error of recognizing the same clothes is generated even though the clothes have different colors. Accordingly, the clothes detecting apparatus 100 according to the present invention is characterized in that clothing is searched using both the color information of the clothes and the texture information of the clothes through the color similarity calculation unit 130 and the texture similarity calculation unit 140 .

On the other hand, even if both the color information of the clothes and the texture information of the clothes are used, it may happen that the completely identical clothes can not be found. This is because texture information that is not found as a feature point is missing and may be recognized as the same clothes despite being different clothes. Accordingly, the clothes detecting apparatus 100 according to the present invention can utilize the information of the person wearing the clothes to improve the recognition rate of the clothes. Information about the person can be acquired through the person information acquiring unit 150. [

Using information about a person to find clothes can further reduce the error rate of mistaking the clothes and can further shorten the time for searching / recognizing the clothes. In addition, authoring time, which is a key issue of video authoring tools, can be further shortened. This is possible because the garment is recognized based on the recognized face without searching the garment area for all the areas.

The power supply unit 170 serves to supply power to all components of the clothes detecting apparatus 100 and the control unit 180 serves to control the overall operation of all the components of the clothes detecting apparatus 100.

Hereinafter, a garment detecting method of the garment detecting apparatus 100 will be described with reference to Fig.

First, the face recognition unit 110 recognizes a face of a person in the image (S200). Then, the garment region extracting unit 120 extracts a garment area located at the bottom of the face based on the position of the face in the image (S210). At this time, the garment area to be extracted is preferably a region-of-interest (ROI) centered on the user.

In order to match each piece of color information and texture information, the ROI of a portion considered as a stream in each image should be set, and a search for the target ROI should be performed. Basically, it is assumed that the face search in each image is finished. It is also assumed that the human body is already garmented. Under this assumption, the ROI is set by capturing the area under the face at a certain ratio of the face size. However, the user manually sets the ROI manually for the image input by the query. However, the other image to be compared sets the ROI, which is considered as clothing, to be proportional to the face size at a certain portion under the face. Referring to FIG. 3, reference numeral 300 denotes a face of a person shown in the image, and reference numeral 310 denotes an ROI. At least one ROI is extracted.

face recognized image = fr + i

In the above, fr is the face area information, and i is the original image. In other words, it can be seen that the image in which the face of the person is recognized includes the information about the face region of the person in the image itself.

ROI.top = fr.bottom + fr.Height / 3

ROI.bottom = fr.Height × 2

ROI.left = fr.left - fr.Width

ROI.right = ROI.left + fr.Width x 3

Equation (2) relates to a method of setting an area considered as a garment using a face area. According to Equation (2), the lower portion excluding the neck portion is approximately 3 times as wide as the face size (preferably 2.5 to 3.5 times as wide as the face size) 2.5 times the height).

When the garment area, that is, the ROI is extracted, the color similarity calculation unit 130 color matches the query input image from which the ROI is extracted to the comparison target image (S220a). Then, the color similarity degree calculation unit 130 calculates the color similarity degree of the query input image (S221a). If the calculated color similarity value of the query input image is larger than the reference similarity value, the query input image at this time is determined to be similar to the color of the comparison target image.

In the embodiment of the present invention, the color similarity calculation unit 130 converts the RGB color space to the HSI color space to measure the degree of color matching in the garment, and calculates the degree of similarity between the H elements using the histogram intersection method .

Basically, the color space of each frame decoded from the video data is RGB (Red, Green, Blue). However, since it is difficult to extract a color-only part from each RGB element, it is necessary to first convert the image into HSI color space that can be separated into H (Hue, Color), S (Saturation) . Then, color information of the image is analyzed using H which contains color information. The RGB HSI conversion is expressed by Equation (3).

In Equation (3), H has a value between 0 and 360 degrees. However, since the values in parentheses can have acceptable values up to the decimal point, the measurement of similarity between different H values may be difficult. Thus, the value between 0 ° and 360 ° is normalized to an integer value between 0 and 12, thereby increasing the efficiency of the similarity measurement. The conversion method is shown in Equation (4).

In the above, NH is a normalized H value, and n is a normalization step and becomes 12 under the same conditions as above.

(3) and (4) to convert both the query input image including the ROI and the comparison image. Then, the difference of each H value is analyzed, and the color similarity is measured. In order to accurately measure the color similarity, it is desirable to set a predetermined number of blocks without comparing the entire image, find the block having the highest similarity first, and then find the average value of all the blocks. This will be described in more detail as follows.

First, the input image is divided into blocks. Then, each block of the image to be compared is scanned using the divided blocks. Figure 4 is an illustration of a method of matching between blocks. Then, the similarity value of the block in which the maximum similarity degree is obtained for the comparison object image among the scanned blocks is stored. Then, an average value of the similarity values of the stored blocks is obtained, and the overall color similarity is measured.

The similarity of each block is calculated by measuring whether the distribution of H values in the block is similar to each other using the histogram intersection method. The histogram intersection method is shown in Equation (5).

In the above, SI is a histogram of a query input image, and CI is a histogram of an image to be compared.

If the calculated degree of similarity is 50% or more, it is regarded as a similar block. If the number of blocks regarded as pseudo blocks is 50% or more of the total number of blocks, it is judged that the image inputted as the query is similar to the color of the comparison object. This can be expressed by Equation (6).

In the above, Fcm is a color matching evaluation function.

On the other hand, when the garment area is extracted, the texture similarity calculation unit 140 texture-matches the query input image with the comparison object image (S220b). Then, the texture similarity calculation unit 140 calculates the texture similarity of the query input image (S221b). In the present invention, the function of the texture similarity calculation unit 140 may be performed simultaneously with the function of the color similarity calculation unit 130, and may be performed after the function of the color similarity calculation unit 130 is performed.

In the embodiment of the present invention, the texture similarity calculation unit 140 extracts a texture feature vector for comparing and measuring clothing texture in a clothing search, and determines a distance between the texture vectors to measure texture similarity of the clothing.

The garment texture is a morphological pattern of a distinctive cloth. The morphological pattern of the image can be expressed numerically by the feature of the frequency coefficient of the image. Thus, first, the image is converted to the frequency domain, and the texture feature vector is calculated from this frequency domain.

First, DCT is used to transform the frequency domain. The DCT method is as shown in Equation (7). The DCT transform is a discrete transform in which sinusoidal wave and cosine wave function are used as basic vectors. When transforming DCT by dividing the image into 8 × 8 block sizes using DCT, 64 transform coefficients can be obtained. Among these, the coefficients of the portion that reflects the morphological characteristics are extracted to construct the texture feature vector.

In this case, the feature vector is extracted by the coefficients of each direction of the block. In order to reflect morphological characteristics well, the feature vector is extracted by the coefficients in the x-axis direction as shown in FIG. 5 (a) The feature vectors may be extracted using diagonal coefficients or the feature vectors may be extracted using diagonal coefficients. Then, the frequency components included in the extracted feature vectors reflect the morphological change of the query input image in the image of the spatial domain.

In the 8x8 block, nine feature vectors are calculated through Fig. 5 (c) through Fig. 5 (b). In Figures 5 (b) and 5 (c), each numeral indicates the number of a vector. Nine feature vectors are extracted as the sum of the DCT coefficients of each digit position. Equation (8) shows a method for estimating a texture feature vector value.

When the feature vector is calculated by the above-described method, 9 vector values are present for each block in the image divided into the 8 x 8 block size. Then, the block of the image to be compared is scanned using one block of the query input image, and the distance of each of the nine vectors is obtained. The method for obtaining the vector-to-vector distance is shown in Equation (9).

The distance between the texture vectors can be obtained by summing the distances of the respective blocks calculated using Equation (9). If the distance between the calculated texture vectors is within an appropriate distance, the query input image at this time is determined to be similar to the texture of the comparison object. This can be referred to Equation (10).

When the color similarity degree calculation unit 130 and the texture similarity degree calculation unit 140 respectively calculate the color similarity degree and the texture similarity degree of the query input image, the clothing detection unit 160 compares each calculated similarity value with the reference value. Then, the garment detecting unit 160 determines that the color or texture of the query input image is similar to the comparison object (S230a, S230b) if the color similarity degree or the texture similarity degree is not less than the reference value.

The garment detecting unit 160 detects the ROI of the query input image determined as similar after step S230 as a garment image (S240).

Meanwhile, the present invention can also generate garment-related metadata using the garment detection method. Hereinafter, a method of generating garment-related metadata will be described. FIG. 6 is a flowchart illustrating a method of generating garment-related metadata using a garment detection method according to a preferred embodiment of the present invention. This will be described below with reference to FIG.

Metadata typically refers to data that describes or represents the meaning of the data. In the case of garment recognition, metadata means clothing-related data included in images such as color of clothes, texture of clothes, and the like.

First, when a moving image is input (S600), the face region of the person is recognized in the input moving image (S610). Then, the garment region ROI located below the face is extracted based on the recognized face region (S620). Then, the color similarity and texture similarity, which are features of clothes, are calculated in the extracted clothing area ROI (S630). Then, the calculated similarity value is compared with a predetermined reference value to detect the ROI of the query input image determined to be similar. In this process, the garment area for the entire moving image is searched to find an area for the garment having the greatest similarity (S640).

The steps S600 to S640 have already been described, and a description thereof will be omitted here.

If the search for the entire moving image area is not completed (S650), steps S620 to S640 are repeatedly performed. On the other hand, if the search for the entire video area is completed (S650), the ROI is finally collected to generate metadata (S660).

A moving picture editor having a structure capable of adding an image recognition function when the metadata generation method of FIG. 6 is implemented as a system, a method of extracting a representative frame from an image and detecting a face recognized position and metadata And an editor for generating the data. In addition, such a system also performs a function of finding color information and texture information of a garment to be searched at the lower part centered on the face recognition position.

It will be apparent to those skilled in the art that various modifications, substitutions and substitutions are possible, without departing from the scope and spirit of the invention as disclosed in the accompanying claims. will be. Therefore, the embodiments disclosed in the present invention and the accompanying drawings are intended to illustrate and not to limit the technical spirit of the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments and the accompanying drawings . The scope of protection of the present invention should be construed according to the following claims, and all technical ideas within the scope of equivalents should be construed as falling within the scope of the present invention.

The present invention discloses a garment detecting apparatus for recognizing a face of a person in an image and detecting clothing based on the recognized face and a method thereof. INDUSTRIAL APPLICABILITY According to the present invention, it is possible to develop a garment recognition technology having an excellent recognition rate from an image and a high execution speed. In addition, it is possible to greatly shorten the metadata authoring time of the moving image than before.

Garment recognition technology for detecting garments in images can be conveniently used in existing Internet shopping malls, especially clothing-related shopping malls. In addition, metadata generation technology can be used by companies providing various services based on moving image objects. The clothes detecting method and the metadata generating method according to the present invention are expected to secure competitiveness in the related industry due to excellent recognition rate, automatic detection, shortening of authoring time due to automatic generation, and the like.

1 is a block diagram schematically showing a clothes detecting apparatus according to a preferred embodiment of the present invention,

2 is a flowchart illustrating a garment detection method according to a preferred embodiment of the present invention,

FIG. 3 is an example of setting a garment area using a face recognized in an image,

4 is an exemplary diagram of a method of inter-block matching,

FIG. 5 is an exemplary diagram of the characteristics of DCT coefficients and a method of extracting texture feature vectors,

FIG. 6 is a flowchart illustrating a method of generating garment-related metadata using a garment detection method according to a preferred embodiment of the present invention.

Description of the Related Art

100: Appearance detecting apparatus 110: Face recognition unit

120: garment region extracting unit 130: color similarity calculating unit

140: texture similarity calculation unit 150:

160: Clothing detecting unit 170:

180: control unit 310: ROI

Claims (14)

A face recognition unit for recognizing a face of a person in a query input image inputted as a query; A person information acquiring unit for acquiring person information indicated by the recognized face; A garment region extracting unit for extracting a garment region from the query input image based on the recognized face; A color similarity calculation unit for color matching the extracted garment area to a predetermined area of a comparison object image and calculating a color similarity value of the garment area from the color matching; A texture similarity calculation unit for matching the extracted garment area with a predetermined area of a comparison target image and calculating a texture similarity value of the garment area from the texture matching; And Comparing the calculated color similarity value and the calculated texture similarity value with a predetermined reference value to determine whether the color or texture of the extracted garment area is similar to a predetermined area of the comparison object image, And a garment detecting unit for detecting the garment area as a garment image representing the garment, Wherein the clothes detecting unit comprises: And the acquired person information is further used when the garment image is detected. The method according to claim 1, Wherein the garment region extractor sets the face region based on the size of the face and extracts the garment region from a position below the set face region. The method according to claim 1, The color similarity calculation unit may calculate, A color space conversion unit for converting the color space of the frame including the garment area from RGB to HSI; A color information deriving unit for deriving information on a color in consideration of the RGB in the converted image; And A color similarity value calculation unit for comparing the two color information and calculating the color similarity value when the information about the color is derived from the predetermined area of the comparison object image, And an image processing unit for processing the image. The method according to claim 1, The texture similarity calculation unit may calculate, A frequency domain switching unit for converting an image of the garment area into a frequency domain using DCT; A texture feature vector extractor for extracting a frequency domain coefficient from among transform coefficients for each block of the image derived from the conversion and extracting a texture feature vector including the extracted frequency domain coefficients; And A texture similarity value calculating unit that compares the extracted texture feature vectors with texture feature vectors in a predetermined area of the comparison object to obtain distances between corresponding texture feature vectors and then calculates a similarity value by summing the obtained distances; Calculation unit And an image processing unit for processing the image. The method of claim 3, Wherein the color similarity value calculation unit divides the query input image into blocks, extracts at least one block from the divided blocks, and compares the extracted block with a block of the comparison object, And comparing the two pieces of color information. The method according to claim 1, A person information acquiring unit for acquiring information about a person indicated by the recognized face, Further comprising: a detecting unit for detecting the garment in the image. 3. The method of claim 2, The garment region extracting unit extracts, A face area size determination unit for determining a horizontal size and a vertical size of the set face area; A garment area detector for detecting the garment area having a width of 2.5 to 3.5 times the width of the face area at a position below the face area and a height of 1.5 to 2.5 times the vertical size of the face area; And an image processing unit for processing the image. The method of claim 3, The color information derivation unit derives a color value in the garment area including a R (Red) component value, a G (Green) component value, and a B (Blue) component value as information on the color, Is normalized to a value in the range of 0 to 12. 5. The method of claim 4, Wherein the distance calculator calculates a distance between the texture feature vectors using nine texture feature vector values that exist for each block. (a) recognizing a face of a person in a query input image input by a query; (b) extracting a garment area from the query input image based on the recognized face; (c) color matching the extracted garment area to a predetermined area of a comparison image, and calculating a color similarity value of the garment area from the color matching; (d) texture matching the extracted garment area to a predetermined area of a comparison object image, and calculating a texture similarity value of the garment area from the texture matching; (d ') acquiring information about a person indicated by the recognized face; (e) comparing the calculated color similarity value and the calculated texture similarity value with a predetermined reference value to determine whether the color or texture of the extracted garment area is similar to a predetermined area of the comparison object image, Detecting the garment area as a garment image representing the garment, The step (e) And when the clothes image is detected, information on the acquired person is further used. 11. The method of claim 10, The step (c) (ca) transforming the color space of the frame including the garment area from RGB to HSI; (cb) deriving information on the color in consideration of the RGB in the transformed image; And (cc) when information on a color is derived from a predetermined area of the comparison target image, calculating the color similarity value by comparing the two color information ≪ / RTI > 11. The method of claim 10, The step (d) (da) converting the image for the garment area into frequency domain using DCT; (db) extracting a frequency domain coefficient among transform coefficients for each block of the image derived from the transform and extracting a texture feature vector including the extracted coefficients for each frequency domain; (dc) comparing the extracted texture feature vectors with texture feature vectors in a predetermined region of the comparison object to obtain distances between corresponding texture feature vectors; And (dd) calculating the texture similarity value by summing all the obtained distances ≪ / RTI > 11. The method of claim 10, The step (b) (ba) setting the face region based on the size of the face; And (bb) extracting the garment area at a position below the set face area ≪ / RTI > delete

Images