KR101071911B1 - Method for creation 3 dimensional image - Google Patents

Abstract

The present invention relates to a method for converting two-dimensional image data into three-dimensional image data, and more particularly, to a method for producing three-dimensional image data having depth from the two-dimensional image data.
To this end, the method for converting 2D image data into 3D image data includes extracting an object from the displayed 2D image data, and quantitatively extracting the extracted objects by tracking the vanishing point of the camera and the image of the camera that captured the image. Assigning to a depth layer of one of a plurality of depth layers having a depth value, and generating a distance recognition map representing a texture of an object allocated to each depth layer based on concept modeling, wherein the plurality of depth layers Image data (thin-depth image data) located at a short distance among two-dimensional image data, which are divided into the first to fourth layers, are allocated to the first layer, and image data (deep image data) located at a far distance. Assigns to the fourth layer to implement a final stereoscopic image. In addition, a key distance recognition map, which is a final stereoscopic image, is generated by assigning an identifier having the same depth information, and a plurality of similar frames are generated using the generated distance recognition map.

Description

Method for creation 3 dimensional image {Method for creation 3 dimensional image}

The present invention relates to a method for converting two-dimensional image data into three-dimensional image data, and more particularly, to a method for producing three-dimensional image data having depth from the two-dimensional image data.

With the development of digital technology, three-dimensional image technology has been widely used. 3D imaging technology is a technology for representing a more realistic image by giving information about the depth of the 2D image.

Since the human eye is separated by a predetermined distance in the horizontal direction, the two-dimensional images seen by the left and right eyes are different from each other. This is called binocular disparity. The brain fuses two different two-dimensional images to produce a three-dimensional image with perspective and realism.

Three-dimensional image technology has a technique for producing three-dimensional image data reflecting the information on the depth, or to generate the three-dimensional image data by converting the two-dimensional image data, the research on both technologies are being done together.

In order to generate 3D image data using 2D image data, information indicating a distance between a photographing target and a camera is required. One example of such information is depth information. The depth information indicates how far the camera and the object represented by the pixel are for each pixel constituting the 2D image data.

In general, the following three methods may be used to acquire depth information.

First, depth information may be obtained by analyzing a shape of a photographed object. This method is economical in that it uses one piece of two-dimensional image data. However, there is a problem that it is difficult to put practical use because it is difficult to implement a device or method for analyzing the shape of the object.

Second, depth information is obtained by analyzing two or more two-dimensional image data obtained by photographing the same object from different angles. This method is widely used in that it is easy to implement. However, in order to photograph the same object from different angles, the photographing means (for example, a camera) must have a plurality of optical systems having different light paths. Since the optical system is an expensive article, it is uneconomical for the imaging means to have two or more optical systems.

Third, three-dimensional image data may be obtained by analyzing two or more two-dimensional image data photographing the same object. For example, in the paper "Simple range cameras based on focal error" (A. Pentland, S. Scherock, T. Darrell, and B. Girod), the paper analyzes the focused and unfocused images to obtain depth information. The configuration is disclosed.

Equation 1 below is an equation used to obtain depth information using two or more two-dimensional image data extracted from the above-mentioned paper. Equation 1 shows an example of obtaining 3D image data using two or more image data.

[Equation 1]

In Equation 1, f is a focus value of a camera lens, D is a distance between an image plane located between the camera and the lens, and r is a radius of a portion where a photographing target is blurred due to out of focus. In addition, k is a conversion constant, and fnumber is a value of f of the camera divided by the focal length of the camera lens divided by the lens aperture. Among these, other values except for r are related to the physical conditions of the camera and can be acquired at the same time as the photographing. Therefore, when the r value is obtained from the captured image, depth information may be obtained.

The focal value f of the camera lens used in Equation 1 indicates physical characteristics of the lens, and thus the focal value f cannot be changed while photographing a photographing target with the same camera. However, the focal length, which will be described later, adjusts the distance between the lenses so that the photographing target is focused, so that the focal length can be changed while the photographing target is captured by the same camera.

As described above, there have been various technologies for converting two-dimensional image data into three-dimensional image data. However, there is a need for various methods to realize live three-dimensional (three-dimensional) images by placing each object naturally at the correct distance.

The problem to be solved by the present invention is to propose a method for producing three-dimensional image data having a depth from the two-dimensional image data.

Another problem to be solved by the present invention is to propose a method for realizing a lively three-dimensional image by allowing each object constituting the two-dimensional image data to be naturally located at the correct distance.

To this end, the method for converting 2D image data into 3D image data includes extracting an object from the displayed 2D image data, and quantitatively extracting the extracted objects by tracking the vanishing point of the camera and the image of the camera that captured the image. Assigning to a depth layer of one of a plurality of depth layers having a depth value, and generating a distance recognition map representing a texture of an object allocated to each depth layer based on concept modeling, wherein the plurality of depth layers Image data (thin-depth image data) located at a short distance among two-dimensional image data, which are divided into the first to fourth layers, are allocated to the first layer, and image data (deep image data) located at a far distance. Is assigned to the fourth layer to implement the final stereoscopic image. And the body features a vector image tracking method that can be applied to a frame similar to the average 100 sheets.

In general, the 3D stereo conversion process requires a large number of people and a lot of work time, but according to the 3D stereoscopic image conversion technology, the number of people and working time can be greatly reduced, so that more 3D stereoscopic content can be generated during the same time. Can be. In addition, it is possible to work more finely by using the technique proposed in the present invention to parts that are difficult to do manually (ex: the end of the hair, etc.) can improve the quality of the content.

In addition, by combining know-how that cannot be done by the operator's program and techniques that cannot be done by hand, it is possible to generate high-quality stereoscopic converted image contents with improved work speed.

1 is a flowchart illustrating a process of converting two-dimensional data into three-dimensional data according to an embodiment of the present invention.
2 illustrates a process of converting 2D image data into 3D image data according to an embodiment of the present invention.
3 illustrates an example of generating a Bezier curve, a second Bezier curve, and a third Bezier curve using a Bezier curve algorithm.
4 illustrates a process of restoring a target region using a texture cinesis according to an embodiment of the present invention.
5 illustrates an example of providing a depth value to an extracted object according to an embodiment of the present invention.

The foregoing and further aspects of the present invention will become more apparent through the preferred embodiments described with reference to the accompanying drawings. Hereinafter will be described in detail to enable those skilled in the art to easily understand and reproduce through this embodiment of the present invention.

The present invention proposes a method of realizing a three-dimensional image lively by converting two-dimensional image data to a pixel unit using a multi-layer / distance recognition map to have a sense of depth, so that each object is naturally located at the correct distance.

In general, 100 similar images can be produced using one key distance recognition map, and the present invention provides a method for producing a key distance recognition map, which is the basis for producing similar images, using a multi-layer production technique. Suggest.

In general, the most difficult process of converting 2D image data into 3D stereoscopic image data is designing a sense of space, and the present invention has the same sense of space as an image taken by an actual stereoscopic camera, Since the same volume as the real object can be applied, the same image as the real 3D image data can be obtained. In addition, since the key distance recognition map can be used for all related scenes, the time for converting 2D image data into 3D image data can be shortened.

1 is a flowchart illustrating a process of converting two-dimensional data into three-dimensional data according to an embodiment of the present invention. Referring to FIG. 1, a process of converting two-dimensional data into three-dimensional data according to the present invention is performed as follows.

In operation S100, the image conversion apparatus performs an object extraction step of extracting an object from 2D image data.

In operation S102, the image converting apparatus assigns a depth layer to the extracted object. That is, the object extracted by tracking the vanishing point of the camera and the image of the camera that captured the image is allocated to one depth layer of a plurality of depth layers having a quantitative depth value.

In operation S104, the image conversion apparatus generates a distance recognition map using a concept modeling technique on the object to which the depth layer is assigned.

2 illustrates a process of converting 2D image data into 3D image data according to an embodiment of the present invention. Referring to FIG. 2, the 2D image data generates 3D image data through an object extraction step, a layer assignment step, and a distance recognition map generation step.

Hereinafter, a process of generating 3D image data from the 2D image data mentioned in FIG. 1 will be described in sequence.

First, an object extraction step of extracting an object from 2D image data will be described.

The object extraction step utilizes path generation and editing techniques and hole reconstruction techniques for mask generation and correction.

The path creation and editing method is a method of generating a natural path using a Bezier curve and a contour tracking function to automatically generate the contour and path handle of an object, and the path using the handle information of the path. Includes animation techniques.

Bezier curves occupy significant positions in numerical analysis from n-first order curves derived from n points. In particular, third-order Bezier curves are used to create smooth curves in PostScript fonts, METAFONT, GIMP, etc., and second-order Bezier curve algorithms are used for true-type fonts.

3 illustrates an example of generating a Bezier curve, a second Bezier curve, and a third Bezier curve using a Bezier curve algorithm.

Path animation is an AnimationTimeline format that takes a PathGeometry as input. Instead of setting the From, To, or By properties as in the From / To / By animation, or using keyframes as in keyframe animations, you define a geometric path and use it to set the PathGeometry property of the path animation. As the path animation progresses, it reads x, y, and angle information from the path and uses that information to generate the output.

In addition, the present invention proposes a hole restoration technique to extract an object. The hole reconstruction technique is a texture synthesis technique that generates a texture of a target region using a similar pattern of a peripheral region.

4 illustrates a process of restoring a target region using a texture cinesis according to an embodiment of the present invention.

Next, the depth layering process will be described. In the depth layer assigning process, the extracted objects are aligned with a depth layer having a quantitative depth value. This will be described in detail below.

Arrange objects with similar depth values on the same layer. In general, each layer has a standardized depth value, and each object gives a reference depth value and a label having the same identifier.

Therefore, the present invention has an advantage that if only the depth value given to the level is modified, the depth value of the corresponding objects is also modified, which is advantageous when a plurality of people work together.

5 illustrates an example of providing a depth value to an extracted object according to an embodiment of the present invention. Referring to FIG. 5, the depth value is divided into four layers including a front layer, a convergence layer, a first rear layer, and a second rear layer. Although FIG. 5 illustrates an example divided into four layers, the number of layers may vary according to a user's setting. For example, the number of layers may be composed of at least two, and in order to have various depth values, it is preferable to increase the number of layers.

That is, as the number of layers increases, the depth value of the corresponding object may be subdivided. When the number of layers decreases, the time for assigning the depth value of the corresponding object may be shortened. Therefore, the number of layers can be adjusted according to the difficulty of converting a 2D image into a 3D image.

Finally, the process of generating a street recognition map will be described. As described above, the distance recognition map uses a concept modeling technique to express the texture of objects having depth values. Conceptual modeling technique means orderly organization of selected components to describe phenomena, and the components of conceptual modeling are concepts as pictures reflecting assumptions made by philosophy, values and professional practice. The present invention generates a distance recognition map using this conceptual modeling technique. That is, as shown in FIG. 2, a specific part of an object having the same depth value is used to have at least two different depth information by using a conceptual modeling technique to express the texture. A specific part has different depth information and generates a distance recognition map that can recognize different distances in the extracted object.

The virtual image may be generated using the multi-layer / distance recognition map generated in this way, or the multi-view image of more than 3 viewpoints may be generated as well as the 2-view stereoscopic image using the multi-layer / distance recognition map rendering technique. Can be. In addition, not only the left / right movement of the objects but also the rotation information of the target according to the viewpoint can be expressed, thereby making it possible to generate more realistic and accurate stereoscopic images.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the scope of the present invention .

S100: Object Extraction Step S102: Depth Layer Assignment Step
S104: distance recognition map generation step

Claims (7)

In the method for converting two-dimensional image data into three-dimensional image data,
Extracting an object from the displayed two-dimensional image data;
Allocating the extracted objects by tracking a viewpoint and a vanishing point of an image of the camera to which the image is taken, to one depth layer of a plurality of depth layers having a quantitative depth value;
Allocating at least two different depth information to express the texture of the object allocated to each depth layer;
And extracting depth information of an object allocated to each depth layer to generate a distance recognition map capable of recognizing different distances within the object.
The method of claim 1, wherein the extracting of the object comprises:
3D image data conversion method characterized by extracting an object using a Bezier curve algorithm.
3. The depth layer of claim 2, wherein the plurality of depth layers are divided into first to fourth layers, and image data (thin depth image data) located at a short distance of the two-dimensional image data is allocated to the first layer, 3. The method of claim 3, wherein the located image data (depth image data) is allocated to the fourth layer.
The method of claim 3, wherein the object allocated to each layer is assigned an identifier having the same depth information.
The method of claim 4, wherein the depth information of the object allocated to each layer is changed by modifying an identifier having depth information.
In the method for converting two-dimensional image data into three-dimensional image data,
Extracting an object from the displayed two-dimensional image data;
Allocating the extracted objects by tracking the vanishing point of the image and the viewpoint of the camera photographing the image to a depth layer of a plurality of depth layers having a quantitative depth value;
Allocating at least two different depth information to express the texture of the object allocated to each depth layer;
And extracting depth information of an object allocated to each depth layer to generate a distance recognition map capable of recognizing different distances within the object.
A plurality of depth layers are divided into first to fourth layers, and image data (thin depth image data) located at a short distance among two-dimensional image data is allocated to the first layer and image data (depth located at a far distance) Deep image data) is assigned to a fourth layer.
The method of claim 6, wherein the object allocated to each layer is assigned an identifier having the same depth information to generate a key distance recognition map, which is the final stereoscopic image, and to generate a plurality of similar frames using the generated distance recognition map. 3D image data conversion method.

Images