KR100387236B1 - Method and apparatus for generating caricature image - Google Patents

Abstract

Disclosed are a caricature image generating apparatus and method capable of automatically generating a caricature image by detecting and analyzing information on each part of a face in a face image. The caricature image generating apparatus includes a preprocessing unit that preprocesses and normalizes a face image, a face feature detection unit that detects first feature information that is feature information of each part of the face from the normalized face image, and A data storage for pre-storing the second feature information and the models for each of the models and the models, and a feature information matching unit for comparing and matching the first feature information with the second feature information and outputting the result as matching information. And a caricature synthesis unit configured to read corresponding models among models stored in the data storage unit in response to the matching information, synthesize the read models, and output the synthesized result as a caricature image.

Description

Caricature image generating device and method {Method and apparatus for generating caricature image}

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to image synthesis, and more particularly, to a caricature image generating device and method for generating a caricature image obtained by synthesizing a face image by exaggerating or dimming a feature of each part of a face.

As image processing technology has developed, various technologies for photographing and using images have been developed and are under development. Such image processing technology is used in a system for identifying and authenticating a user by detecting and analyzing an input face image and a system for synthesizing a face image. An imaging technique for generating an exaggerated or simplified face image based on the generated image data has been developed.

As one of these imaging techniques, the research system of the National Meeting of Engineers in Tokai Branch, published in 1998, and Chunkyo University, Japan, published in IPSJ Vol134, No10, pages 2106-2116, published in August 1993. An example is PICASSO. The PICASSO system generates an exaggerated face image by targeting a 2D image or a 3D image. Among them, the two-dimensional system compares and analyzes the difference between the data on the average face and the input face data, and generates a caricature of the single line image through an exaggerated cartoon representation of the difference signal. Here, an average face means the face obtained by synthesizing the face which a specific class | group has. In addition, the 3D system detects eyes, nose, mouth, etc. by inputting a range image using depth information, calculates the difference with the average face as in 2D, and generates an exaggerated face image. do.

As another imaging technique, as disclosed in US Patent No. 5,960,099 entitled "System and Method for Creating a Digitized Likeness of Persons," two facial images, front and side, are used for facial features, namely There is a technique of estimating the position of each part through statistical data and generating a portrait with a changed expression.

As another image technology, as disclosed in US Patent No. 5,990,901 entitled "Model Based Image Editing and Correction", a pose of an image is externally specified, and then a specified pose-specific model and an externally input image are synthesized. There is a technique for detecting a desired partial image. This technique transforms the original image into the shape desired by the user by manually assigning all control points, and is widely applied for plastic surgery.

The conventional image processing techniques described above do not represent the original image as it is, but generate an exaggerated or blurred image from the original image, and have the following problems.

First, since the aforementioned PICASSO system exaggerates and emphasizes the difference from the average face image, the PICASSO system cannot fully reflect the features of the original image in the exaggerated and emphasized image. Second, the technique disclosed in U.S. Patent No. 5,960,099 cannot accurately detect the facial features of the original image because the facial features are estimated through statistical data when making portraits with changed facial expressions using the front and side face images. Has Finally, the technique disclosed in U.S. Patent No. 5,990,901 generates a face image in a shape desired by a user by manually setting a control point, so that the features of the original image cannot be sufficiently reflected in the generated face image, but also manually. It is troublesome because the work is done.

As a result, the above-described conventional image processing techniques generate another image based on the original image, but are not suitable for generating an exaggerated or blurred image that sufficiently reflects the facial features of the original image and cannot automatically perform the operation. There is also a problem.

An object of the present invention is to provide a caricature image generating apparatus capable of automatically generating a caricature image by detecting and analyzing feature information of each part of a face from a face image.

Another object of the present invention is to provide a caricature image generation method capable of automatically generating a caricature image by detecting and analyzing feature information of each part of a face in a face image.

1 is a block diagram of a caricature image generating apparatus according to the present invention.

2A is a diagram for describing a process of detecting a lip using an active shape model technique.

3A to 3C illustrate detailed models of eyes and eyebrows stored in a data storage unit.

4A-C are diagrams showing models for the front and rear heads.

5a to c are diagrams showing examples of a hairstyle synthesized by combining the model of the front head and the model of the back head.

FIG. 6 is a diagram illustrating a process of matching a head shape according to head region segmentation from an input face image.

7 is a diagram illustrating a synthesis example of a caricature image.

8 is a flowchart for explaining a caricature image generation method according to the present invention.

In order to achieve the above object, the caricature image generating apparatus according to the present invention for generating a caricature image of the face from the face image obtained by capturing the face, the face image by a face search method using a Gaber Filter Response (Gaber Filter Response) Pre-processing to normalize and output a normalized face image, and first feature information, which is characteristic information about each part of the face including the nose and the head, from the normalized face image. A face feature detection unit for detecting using a shape model), second feature information which is feature information for each model of each part of the face, a data storage unit for pre-storing the models, and the first feature information A feature information matching unit and the data for comparing and matching with the second feature information and outputting the result as matching information; And a caricature synthesis unit configured to read corresponding models among models stored in a storage unit in response to the matching information, and to synthesize the read out models as a caricature image, wherein the feature information matching unit is configured to output the caricature image. A Fourier Descriptor for boundary information is calculated from the feature information, and the matching information is generated in response to the calculated result.

In order to achieve the above object, the caricature image generating method according to the present invention for generating a caricature image of the face from the face image obtained by capturing the face, the face search method using a Gaber Filter Response (Gaber Filter Response) Normalizing a face image by preprocessing; obtaining first feature information, which is feature information on each part of the face, from the normalized face image by using an active shape model technique; Obtaining matching information by comparing and matching the models of the site and the pre-stored second feature information, which is the feature information of each of the models, with the first feature information and matching the corresponding models of the models to the matching information. And correspondingly selecting and synthesizing the selected models to generate the caricature image. In the obtaining of the matching information, a Fourier Descriptor for boundary information is calculated from the first and second feature information, and the matching information is obtained using the calculated result.

Hereinafter, the configuration and operation of a caricature image generating apparatus according to the present invention will be described with reference to the accompanying drawings.

1 is a block diagram of a caricature image generating apparatus according to the present invention, which includes a preprocessing unit 10, a facial feature detection unit 12, a feature information matching unit 14, a data storage unit 16, and a caricature synthesis unit 18. It includes.

The preprocessing unit 10 illustrated in FIG. 1 preprocesses and normalizes a face image input through the input terminal IN, generates normalized face image data, and outputs the normalized face image data to the face feature detection unit 12. Here, the input face image is obtained by capturing a face through an image photographing device (not shown) such as a camera.

The preprocessor 10 may use, for example, a Gabor Filter Response for the above-described function. That is, the preprocessing unit 10 detects a face region from an input face image by detecting a gabor filter response, detects a center point of two eyes, and creates a normalized face image based on the normalized face image. Perform Skin Color Filtering.

Meanwhile, the facial feature detector 12 detects features of eyes, eyebrows, noses, lips, facial contours, heads, and the like which are parts of the face from the normalized facial image output from the preprocessor 10, and detects the detected features. Is output to the feature information matching section 14 as first feature information. For this purpose, for example, the facial feature detector 12 may use an Active Shape Model (ASM) technique. The ASM technique is a technique for extracting various types of representative models and characteristics of morphological changes in each part of the face.

In order to help the understanding of the present invention, the facial feature detection unit 12 uses the ASM technique to extract various types of representative models of the lips and the features of the morphological changes using the ASM technique. If described with reference to:

FIG. 2A is a diagram illustrating a process of detecting lips using the ASM technique, FIG. 2A illustrates a normalized face image, FIG. 2B illustrates a learning pattern process for obtaining a representative model of the lips, and FIG. 2C Is a diagram showing the relationship between the average coordinates of the representative model 20 of the lips obtained through the learning pattern and the target lips, and FIG. 2D is a diagram showing the detection of the lips.

First, various types of lip shapes as shown in FIG. 2B are used as learning patterns to obtain representative models of lips, average position coordinates, and features of morphological changes of lips. In this case, when the normalized face image shown in FIG. 2A is input from the preprocessing unit 10, the facial feature detection unit 12 applies the lip representative model 20 to previously obtained average coordinates, Find the lip contour of the input face considering the features. Therefore, even if the skin color and the lip color are similar and the boundary of the lips, that is, the lip contour is unclear, the lip contour can be well detected as shown in FIG. 2D through statistical information obtained through learning. In this manner, feature information about other parts of the face may be detected, and the detected feature information is output to the feature information matching unit 14 as first feature information.

The data storage unit 16 shown in FIG. 1 includes a plurality of models of each part of the face, that is, a head, a face, an eye, an eyebrow, a nose, a mouth, and the like, and second feature information which is feature information about the models. Are stored in the form of a database (DB). The basic principle of DB composition is to divide the types that appear most for each face area by rank and construct the model to reflect the characteristics of the type and to look beautiful.

For example, among the models stored in the data storage unit 16, the eye and eyebrow models that have the greatest influence on the impression of the person are as follows.

3A to 3C illustrate detailed models of eyes and eyebrows stored in the data storage unit 16.

In each of FIGS. 3A to 3C, eyes that appear most commonly are classified into seven types. FIG. 3A shows the models of the eye with normal tail, FIG. 3B shows the models of the eye with raised tail, and FIG. 3C shows the models of the eye with lower tail, respectively, in FIGS. 3A-C, in columns A and B And C represent normal-sized eyes, double eye, and small eyes, respectively. Therefore, 7 * 3 * 3 = 63 kinds of eye models may be configured as a DB. However, as shown in Figs. 3A to 3C, the data storage unit 16 stores only 54 detailed models in DB except for 9 cases which do not appear to be actual. On the other hand, the eyebrows are classified into 48 types. First, nine different shapes appearing according to men and women are divided into DBs with raised tails, lowered types, and normal types, and then divided into two types according to the thickening of the eyebrows. .

As another example, looking at the model for the head of the model stored in the data storage unit 16 as follows.

Figures 4a-c show models for bangs and foreheads, Figure 4a shows models for the forehead, Figure 4b shows the models for the forehead, and Figure 4c shows for the foreheads. Represent each model.

The front and back hairstyle models shown in FIGS. 4A-C are stored in the data storage 16 and may be extracted from a particular group, ie, teenager, 20 and 30 hair styles.

As another example, among the models stored in the data storage unit 16, models (not shown) for the face shape may be largely composed of three types, slim type, normal type, and chubby type. At this time, the square jaw or the pointed jaw drop is excluded from the models to be stored. In addition, the models (not shown) for the lips to be stored in the data storage unit 16 may be configured in about 10 or more depending on the shape, and is again composed of a universal lipstick color and four natural colors into a DB.

As a result, the data storage unit 16 pre-stores the models for each part of the face as described above and the second feature information to be used for matching each of the models, and stores the second feature information as the feature information matcher 14. And the caricature synthesis unit 18 are output. The configuration and contents of the models and the second feature information stored in the data storage unit 16 may be updated according to a user's request.

Meanwhile, the feature information matching unit 14 analyzes the first feature information detected and provided by the face feature detection unit 12, reads the second feature information stored in the data storage unit 16, and analyzes the first feature information. And match, and output the result to the caricature synthesis unit 18 as matching information. To this end, for example, the feature information matching unit 14 calculates a Fourier Descriptor for the boundary information, and thus, the first feature information for each face and the second feature information stored in the data storage unit 16. Can be matched. When such a match is made, the feature information matching unit 14 outputs the matching information, which is a result of the matching, to the caricature synthesis unit 18.

Meanwhile, the caricature synthesis unit 18 shown in FIG. 1 reads out corresponding models to be synthesized for the caricature image among the models stored in the data storage unit 16 according to the matching information output from the feature information matching unit 14. The synthesized models are synthesized, and the synthesized result is output as a caricature image through an output terminal OUT.

The process of synthesizing the head portion of the caricature image by the caricature synthesis unit 18 is as follows.

FIGS. 5A to 5C are diagrams showing examples of a hairstyle synthesized by combining the model of the front head shown in FIG. 4A or C and the model of the back head shown in FIG. 4B.

The caricature synthesis unit 18 reads one of the front head models shown in FIG. 4A or C and one of the back head models shown in FIG. 4B from the data storage unit 16 using matching information, The read models may be synthesized to generate a head model of a caricature image as shown in FIGS. 5A to 5C. In detail, a process of generating the head model of the caricature image is described as follows.

FIG. 6 is a diagram illustrating a process of matching a head shape according to head region segmentation from an input face image.

Referring to FIG. 6, the preprocessing unit 10 illustrated in FIG. 1 inputs and normalizes the face image 30 through the input terminal IN, and converts the normalized face image 40 into the face feature detection unit 12. Output At this time, the face feature detector 12 inputs the normalized face image 40 to detect first feature information about the head and the eyes, and outputs the detected first feature information to the feature information matcher 14. do. The feature information matching unit 14 detects matching information by matching the second feature information stored in the data storage unit 16 with the first feature information thus obtained. In this case, the caricature synthesizing unit 18 performs a hair region segmentation 50 using a region growing method when synthesizing the head portion of the caricature image. In other words, the seed for the region expansion is found through histogram analysis of the hair candidate region, and the hair is segmented through the region expansion. Next, the caricature synthesizing unit 18 uses the divided head region as a template and performs a template matching with the head models shown in FIGS. 4A to C stored in the data storage unit 16 to match the head. Generate a type model 60.

7 is a diagram illustrating a synthesis example of a caricature image.

The caricature synthesizing unit 18 post-processes the caricature synthesized by the above-described method, and completes the synthesis of the caricature image by adjusting the shape of the face, deformation, skin color, and wearing glasses, similarly to the shape of the actual face. For example, the completed caricature image as shown in FIG. 7 is output through the output terminal OUT.

Hereinafter, a method for generating a caricature image according to the present invention will be described with reference to the accompanying drawings.

FIG. 8 is a flowchart illustrating a method for generating a caricature image according to the present invention, comprising: detecting first feature information (steps 70 and 72) and first feature information and a first feature after preprocessing an input face image; And synthesizing the caricature image using the feature information (steps 74 and 76).

First, as described above, the preprocessing unit 10 preprocesses and normalizes a face image obtained by capturing a face in an external image capturing apparatus (step 70). After operation 70, the facial feature detector 12 detects first feature information, which is feature information corresponding to each part of the face, from the normalized face image obtained in operation 70, using the above-described ASM technique. step).

After operation 72, the feature information matching unit 14 analyzes the first feature information obtained in operation 72 and calculates a Fourier descriptor for the boundary information, thereby obtaining the second feature, which is feature information for each of the pre-stored parts of the face. Matching information is obtained by comparing and matching the feature information with the first feature information of the analyzed face (step 74). After the 74 th step, the caricature synthesizing unit 18 obtains the caricature image by synthesizing models of each part of the face selected from the data storage unit 16 according to the matching information obtained in the 74 th step (step 76).

Details of each of the seventy-seventh through seventy-seventh steps are the same as those of the respective parts 10, 12, 14, and 18 shown in FIG.

As described above, the caricature generating apparatus and method according to the present invention match the face image of the original face actually photographed with the models of each part of the face previously stored, and synthesize the caricature through the original face, thereby Caricature images that fully reflect the characteristics of the image can be obtained. In addition, since the caricature image close to the original image can be automatically generated from the input face image, the caricature can be freely stored because the models and the second characteristic information thereof can be stored in the data storage unit 16 as desired by the user. There is also an effect that can be synthesized.

Claims (12)

In the caricature image generating apparatus for generating a caricature image of the face from the face image obtained by imaging the face, A preprocessing unit which preprocesses and normalizes the face image by a face searching method using a Gaber Filter Response, and outputs a normalized face image; A face feature detection unit for detecting first feature information, which is feature information on each part of a face including a nose and a head, using an active shape model from the normalized face image; A data storage for pre-storing the second feature information which is feature information for each of the models of each part of the face and the models; A feature information matching unit which compares and matches the first feature information with the second feature information, and outputs the result as matching information; And A caricature synthesis unit configured to read corresponding models among the models stored in the data storage unit in response to the matching information, synthesize the read models, and output the synthesized image as a caricature image; The feature information matching unit calculates a Fourier Descriptor for boundary information from the first and the second feature information, and generates the matching information in response to the calculated result. Device. delete delete The apparatus of claim 1, wherein the facial feature detector A caricature image generating device for dividing a hairstyle by using a region expansion method through histogram analysis, and detecting feature information of the hairstyle among the first feature information from the segmented result. The method of claim 1, wherein the data storage unit stores the feature information for each of the models for the front and rear hairstyles and the front and rear hairstyles of each part of the face, The caricature synthesis unit synthesizes the model for the front head and the model for the rear head read out from the data storage in response to the matching information, and outputs the synthesized result as the head portion of the caricature image. Caricature image generating device, characterized in that. delete In the caricature image generation method for generating a caricature image of the face from the face image obtained by imaging the face, (A) preprocessing and normalizing the face image by a face searching method using a Gaber Filter Response; (B) obtaining first feature information, which is feature information on each part of the face, from the normalized face image using an active shape model technique; (C) obtaining matching information by comparing and matching the models of each part of the face and the pre-stored second feature information, which is feature information of each of the models, with the first feature information; And (D) selecting corresponding models among the models according to the matching information, and synthesizing the selected models to generate the caricature image; In the step (c), a Fourier Descriptor for boundary information is calculated from the first and second feature information, and the matching information is obtained using the calculated result. Way. delete delete 8. The method of claim 7, wherein the step (b) divides the hairstyle using a region expansion method through histogram analysis, and obtains feature information on the hairstyle among the first feature information using the divided hairstyle. Caricature image generation method characterized in that. delete The method of claim 7, wherein the step (d), And a head portion of the caricature image is obtained by synthesizing the model for the front head and the model for the back head selected according to the matching information.