KR20000063344A - Facial Caricaturing method - Google Patents

Abstract

PURPOSE: A method for automatically creating facial caricature is provided to create the caricature by storing the shape, size, and diverse distinctions as measurement data through an extraction of specific facial distinction and the use of a facial features database, and searching and mapping details corresponding to the person's distinctions. CONSTITUTION: A method for automatically creating facial caricature is composed of the steps of creation, input, extraction, and mapping. Diverse forms of caricature database, which stores each facial detail, are created. When the facial image's data is inputted, the distinctive facial features of the facial image are extracted. The most similar caricature of each facial category is extracted from the database using the extracted distinctive facial feature. The extracted caricature is mapped against the inputted facial image and a caricature similar to the inputted facial image is created.

Description

Automatic face caricature generation method {Facial Caricaturing method}

The present invention relates to the automation of human face caricature generation, and in particular, to extract various features such as shape and size using a detailed feature extraction of the input face image and the mock-up database, rather than drawing existing outlines or using simple size data. The present invention relates to a caricature generation method of generating caricatures by storing numerical data and retrieving and mapping details (such as the number of objects) according to individual characteristics.

Recently, as the price of hardware has spread, the spread of computers and the generalization and activation of the Internet have led to the spread of various services and software in various fields.

In addition, as the speed of the Internet improves, research and technology development for accelerating multimedia services, not just text services, are accelerating on the Internet. Since multimedia services have a greater effect than simple text services, they are currently being used in various fields. This affects mobile communication, and is researching to provide a service that integrates multimedia contents, that is, image, audio, and video, in addition to text services.

This trend has led to the active participation of netizens in recent years due to the spread of the multimedia and to deviate from the provision of one-sided or fragmentary information. Therefore, various efforts have been proposed or researched for an interactive or interactive connection in an interactive movie or various videos.

However, despite the continuous research, the actual multimedia contents are still very insufficient. However, the development of multimedia content is inevitable.

There is a growing interest in characters available in various fields of the multimedia content business. In the virtual space, the use of virtual humans (Avatar) is becoming active, and characters like celebrities or themselves are inserted into various products, and they are also used for the application of the emotional agent that replaces them. In addition, if a caricature is produced by extracting and highlighting or transforming only a few features from the face of a celebrity or an individual, the caricature can be added to various industries as it can give a sense of unity like a real person.

However, despite this possibility, the caricature manufacturing process is mostly done by hand, which causes a lot of manpower and time. Therefore, the present invention is to extract the caricature data of the proof photo to compare the caricature manufacturing process that is made manually with the data of the ready-made eye database, extract a similar car image and automatically develop a system to create individual caricatures Its purpose is to meet these social needs and to prevent waste of money.

Looking at the conventional technology proposed for the automatic caricature generation, for example, the face caricature system proposed in the existing paper, the conventional system uses a fuzzy concept that simplifies the ambiguity and complexity of human thinking. System to generate a caricature.

The conventional system expresses the caricature after comparing the normalized standard face image with the input face image only by comparing the size of the neck and neck to obtain various relative size data. The 21 feature parameters of the standard face image (eye size, eyebrow thickness, nose) The feature information of the input image is extracted using the position, height, face shape, hair length, and the like. Each size data extracted in this way is analyzed by 49 feature terms (big, small, round, thin, tall, short, etc.) to create a caricature by placing similar images on the face. Another system is to express the features of the standard face image (contrast) and face contour by 455 points, and compare the input face image with each point of the standard face image to enlarge and reduce the input image. The system then draws a caricature by contouring the 455 points.

Therefore, the problem of the existing caricature generation system is that the shape of the face is generated by generating the caricature of the face image input through the transformation process based on the fact that the diversity of the face is not reflected because it is generated using various data and the standard face image Because it draws as it is, the effect of the transformation of the photograph rather than the meaning of caricature.

An object of the present invention for solving the above problems is to draw a detailed contour of the input face image and to draw various features such as shape and size by using the database of the head image rather than drawing the existing contour or using a simple size. The present invention provides a method for creating a caricature by generating a caricature by searching and mapping details (such as Lee Mok-Gu's) according to an individual's characteristics.

1 is an exemplary structure diagram of a caricature generation system;

2 is an exemplary face photograph of a model;

3 to 5 is an exemplary view showing a process of extracting the shape of the eyes, nose, mouth, eyebrows, and face of the face using a face photograph,

6 to 8 are exemplary views showing a caricature generation process using the extracted face shape.

A feature of the present invention for achieving the above object is the first process of constructing a caricature database of various forms for each detail (such as yimugubi), and when the data of the facial image is input, such as yimugubi in the face image Extracts the caricature for each sub-item determined to be the most similar among the database constructed in the first process by using the second process of extracting the feature of the detail and the feature of the detail item extracted through the second process. And a fourth process of mapping the caricature extracted through the third process to the face image input through the second process to generate a caricature similar to the input person image.

An additional feature of the present invention for achieving the above object is to further include a user selection mode for the user to select when a plurality of caricature data having the same similarity for a specific item is extracted in the third process .

The above object and various advantages of the present invention will become more apparent from the preferred embodiments of the present invention described below with reference to the accompanying drawings by those skilled in the art.

Hereinafter, exemplary embodiments of the present invention will be described with reference to the accompanying drawings.

1 is an exemplary view showing the structure of a caricature generation system, FIG. 2 is an exemplary view of a face photograph of a model, and FIGS. 3 to 5 are exemplary views illustrating an extraction process for dissatisfaction from a face picture, and FIG. 4 is an exemplary view illustrating a region setting process for extracting the eyes, nose, mouth, eyebrows, and facial shape of the face, and FIG. 4 is an exemplary view of extraction parts for each of eyes, nose, mouth, and eyebrows of the face. 5 is an exemplary diagram of edge extraction to the extracted eunuchsubishi.

In addition, Figures 6 to 8 is an exemplary view showing a caricature generation process using the extracted face shape, Figure 6 is a caricature of the outline of the face, Figure 7 is a caricature for the neck, Figure 8 is of Figures 6 and 7 Complete caricature with synthetic mapping of caricatures.

Looking at the configuration of Figure 1 of the accompanying drawings will be described the relationship between each of the accompanying drawings and Figure 1 to Figures 2 to 8.

1 is not given a reference number, because the system has a concept of a method flow.

First, a segment is provided with a database that stores data of various types of caricatures for each detail (e.g., eyeballs, etc.), and when data of a face image is inputted, segmentation that extracts features of details such as eyeballs from the face image. A facial feature extraction module having functions of over feature information extraction and shape classification is provided.

When the facial feature extraction module extracts the facial feature of the main character of the scanned picture, the extracted image mapping module accesses the data of the most similar caricatures in the database and combines and maps it to the original face image. It is provided.

Therefore, when the data of the part referred to by reference number A in FIG. 1 is a photographic image of a man as shown in FIG. 2, the data processed through the segmentation constituting the facial feature extraction module (reference number B) becomes as shown in FIG.

In this case, the data extracted through the facial feature extraction module is as shown in FIG. 4, and the data edge-processed based on this data is as shown in FIG. 5.

Thereafter, the most similar data is accessed in the database. The shape of the accessed face is as shown in FIG. 6 and the access data corresponding to the object is shown in FIG.

If a plurality of caricature data having the same similarity is accessed in the process of accessing the most similar data in the database based on the data extracted through the facial feature extraction module, all the caricature data accessed by the user are displayed and selected. In this way, the error is minimized while reflecting the user's opinion.

Having looked at the overall concept according to the present invention described above, the technical idea for embodying each concept will be described.

Segmentation is one of the important tasks in the present invention to extract the features of the details, such as portraits of portraits. The ideal result of segmentation is to extract meaningful objects that humans perceive, but until now, no algorithm for extracting meaningful objects to human eyes has been known.

This is the first step to be performed in the module of image processing or analysis. This is also an important part of providing the basic data for the module that needs to be performed. Currently, many image segmentation techniques are available. The types of partial region extraction methods can be classified into edge-based technology, clustering-based technology, and region-based technology.

In the present invention, a color and region feature based partial region extraction method is used, and the procedure to be performed is divided into two steps. The first is the region growing stage and the second is the region merging stage. The region combining step is further subdivided into subdivided and combined noise regions into neighboring regions and recombining according to similarities between the separated subregions.

There is a method of extracting an outline from a circle image and comparing the position and shape of the target image with a binary coordinate point of the outline. There are many theories on how to extract the contours, but it is difficult to find the exact ones. The most basic principle is the method of detecting the point where the slope changes through the first and second derivatives, and many methods have been proposed.

In the present invention, the edge is extracted using the Sobel algorithm. The Sobel calculation method is expressed as Equation 1 below.

[Equation 1]

Since the Sobel calculation method shown in Equation 1 is a calculation formula that is commonly used in the industry, detailed descriptions will be omitted, and outlines are extracted to extract detailed facial features such as each noble eye. Numerical features such as eyes, nose, mouth, eyebrows, etc.

Each feature information includes feature information about a partial region of an input face image. The feature information for the partial region differs according to each component. The eye extracts feature information such as the size of left and right, the angle of inclination of the eyelid, and the up and down eyes. The constructed database classifies the same class. Image classification refers to classifying an image to be searched and assigning it to one of the predefined classes based on the extraction of meaningful features in order to reduce the number of candidates to be searched.

A plurality of shapes extracted from one image are assigned to a plurality of classes according to region feature information. For example, in the case of an eye, when a class is formed by the direction of the eye, there may be various eyes having different sizes in the class of the raised eye. The face image input by using the feature information having such a feature is mapped to a plurality of classes, and the feature information is calculated by searching for images of all classes corresponding to each feature information.

One input face image is classified into classes because the feature information about the shape image of the face and the feature information of each partial region are independently classified into classes, so that each feature of the input face image is searched for. Follow the search to get all the elements of each class.

The classification criteria for the shape information may be expressed by the contour feature of the partial region. Contour information can be subject to any one criterion because subjective interpretation of shape similarity of subregions is possible. Compared to the contour information of the query image, it is classified into the most similar group.

When comparing images, the whole image is calculated in pixel units, so a large amount of calculation is required, and shape feature information is used to calculate similarity. The shape feature information is considered by separating the entire image and the feature information of each partial region. The feature information of the entire face image refers to the face shape obtained through the background and segmentation, and the feature information on each part region varies greatly depending on the eye.

Each caricature image thus extracted is mapped. By calculating the similarity with the input face image, it is a task to detect each detailed attention ratio having a high similarity and map it using the detected image. Mapping uses location information for each detail.

For example, the nose is located in the middle of the face, the eyes are positioned using left and right symmetrical coordinate points with respect to the center of the face, and the position is determined using the distance from the nose in the center of the mouth. Using the detailed caricature extracted from the database, the caricature as shown in FIG. 8 is displayed and mapped and generated through location information.

At this time, after the caricature face image is completed as described in the above description, if there is an exceptionally large, small or one part compared with the statistical numerical data of the normalized standard image, which is made in advance to emphasize the characteristic of the person only, the caricature It will zoom in or out on that part of the image.

For example, if the user's nose is larger than the general public and has a large weight in comparison with other ancestors, the nose of the image obtained by accessing the caricature is enlarged or the nose similar to that of the image is larger than normal. Find and exchange

However, if the images in the database are bitmaps, there is a disadvantage in that the original state of the image is not guaranteed and the image is broken. However, if the data is vector, it is only necessary to enlarge the nose image obtained by accessing it at an appropriate ratio.

That is, in the case of a bitmap or a vector, the feature of the face that only the person has exaggerated. This can be done when the target data is a vector or a bitmap. In this way, caricatures can satisfy the subjective idea of each person or the exaggeration of exaggeration.

In addition, since the present invention has been described as extracting a facial contour, finding a facial shape (illustration) that matches the facial contour, and mapping the found eye, nose, mouth, eyebrows, etc. onto the facial image, the database includes You'll need to have a face shape, but you'll know in advance that you'll be able to choose your outfit, such as glasses or accessories.

In other words, the present invention is applied to the features of the above-described contents, such as haircuts or details, including hair styles, glasses, accessories, and the like.

While the invention has been shown and described in connection with specific embodiments thereof, it is well known in the art that various modifications and changes can be made without departing from the spirit and scope of the invention as indicated by the claims. Anyone who owns it can easily find out.

As described above, when the caricature according to the present invention is automatically generated, it can be used in various fields of the multimedia content business. The use of avatars and caricatures in virtual spaces is becoming active, and characters that resemble entertainers or themselves are inserted into various products, and they are also used for the application of emotional agents in which the characters take their place.

If you use a caricature that resembles a celebrity or yourself, the caricature can give a sense of unity like a real person. By creating a caricature that resembles a user, it is expected to be useful not only in the multimedia field but also in many applications besides the virtual human (Avatar) used in the virtual world of the Internet.

In addition, since talents and caricatures resembling themselves are recently being used in various products, it is expected to spread further by automating them.

Claims (3)

A first step of constructing a variety of types of caricature databases for each of the details (such as the number of objects); A second process of extracting a feature of a detail, such as an eye, from the face image when data of the face image is input; A third step of extracting a caricature for each subitem determined to be the most similar among the databases constructed in the first step by using the feature of the subitem extracted through the second step; And And a fourth process of mapping a caricature extracted through the third process to a face image input through the second process to generate a caricature resembling the input person image. The method of claim 1, After the caricature face image completed through the fourth process is completed, it is exceptionally large or small compared to the statistical numerical data of the normalized standard image, which is made in advance, to highlight the features of the corresponding human face of the source image. And a fifth step of enlarging or reducing the portion of the caricature image if there is a portion. The method of claim 1, And a user selection mode for allowing a user to select when a plurality of caricature data having the same similarity with respect to a specific item is extracted in the third process.