KR20130135606A - A feature based pre-processing method to compensate color mismatching for multi-view video - Google Patents

Abstract

Provided is a feature based multi-view video preprocessing method for correcting an adjacent inter-view color difference caused by the characteristic differences between a camera and an illumination with respect to at least two multi-view videos having different views, whereby the method comprises the steps of: (a) setting a relation function (hereinafter, referred to as a camera characteristic model) for estimating one multi-view video (hereinafter, referred to as a correction view video) from the other multi-view video (hereinafter, referred to as a reference view video); (b) extracting corresponding features from the correction view video and the reference view video; (c) with respect to a feature of the reference view video (hereinafter, referred to as a reference point) and a feature of the correction view video corresponding to the reference point (hereinafter, referred to as a corresponding point), estimating a point by applying the corresponding point to the characteristics model, and obtaining a coefficient of the characteristic model (hereinafter, referred to as a camera characteristic coefficient) for minimizing the difference between the estimated point (hereinafter, referred to as an estimation point) and the reference point; and (d) correcting the correction view video by using the characteristic coefficient. According to the feature based multi-view video preprocessing method, image quality (PSNR) and compression efficiency (bit rate) can be enhanced when encoding a multi-view video image, particularly, the image quality (PSNR) can be improved to as much as approximately 0.5-0.8 dB according to each component in comparison to a conventional cumulative histogram based preprocessing method and the compression efficiency (bit rate) can be reduced to as much as approximately 14%. [Reference numerals] (AA) Start;(BB) End;(S10) Set a camera characteristic model;(S20) Extract corresponding features from two videos;(S30) Obtain a camera characteristic coefficient using the corresponding features;(S40) Correct the video by applying the characteristic coefficient

Description

[0001] The present invention relates to a feature-based multi-view video preprocessing method for color component correction,

The present invention relates to a method of preprocessing a multi-viewpoint video based on feature points for correcting a color difference between adjacent points due to a difference in characteristics between illumination and cameras for at least two multi-viewpoint images having different viewpoints.

In particular, the present invention relates to a method of preprocessing a multi-viewpoint video based on a feature point based on relative feature points between multi-viewpoint images,

With the development of digital video technology in recent years, 3D stereoscopic images, which are more realistic than ever, can be seen without difficulty. As media media are all digitized, stereoscopic 3D stereoscopic images and free-view stereoscopic 3D stereoscopic images can be seen in ordinary households. In addition, related technologies are being actively developed. Image processing using multi-view video (MVC) is a typical example. Unlike conventional single view or stereo image, multi-view video captures multiple scenes and objects by arranging several cameras in a matrix or matrix form, and reproduces stereoscopically using the acquired images to be. Although multi-view video has the advantage of providing viewers with a free viewpoint conversion and a wide viewpoint screen, the amount of information is rapidly increased according to the number of cameras, so that it is limited to various service application fields storing and transmitting multi- . Therefore, it is necessary to develop a technique for efficiently compressing and transmitting a large-capacity multi-view image information.

 Moving picture experts group (MPEG) is working on the standardization work related to compression processing of three-dimensional stereo audio / video signals. In particular, standardization has been proceeding with focus on multi-view video coding, and standardization of MVC has been conducted by JVT (joint video team). Multi-view video coding is implemented by extending H.264 / AVC, a single-view video coding standard [Document 1].

The most important feature of multi - view video coding is to use not only the correlation of the front and rear frames but also the correlation between adjacent points. In other words, the coding efficiency is increased by using inter-view compensation that eliminates spatial redundancy between adjacent view images. However, illumination differences may occur between the images obtained according to the positions of the cameras, or color mismatch problems may occur due to characteristics inherent to the camera sensors. The color discrepancy problem is a phenomenon that the color distribution of the same object changes depending on the viewpoint of the camera used, and may occur even when the camera is used with the same type of camera and the same setting is taken. This lowers the spatial redundancy between adjacent viewpoint images, which increases the cost of inter-view prediction and consequently degrades the overall coding efficiency [2].

In order to solve these problems, a preprocessing method for correcting illumination and color is generally used before encoding. There are three methods for correcting illumination and color between images obtained from cameras at other viewpoints.

The first is a method of correcting color by analyzing the overall characteristics of the image. This is the method of Fecker and Chen [3]. This method is mainly proposed at the beginning of the research for correcting the color of the multi-view image. The color difference is corrected using the overall average brightness histogram of each point and its cumulative histogram. Thus, the method of using the overall characteristics of the image is somewhat difficult to apply to the multi-view image in which the occluded region exists. In other words, the performance of the correction process is not good because the color or brightness information of an object that is visible at one view but not visible at another view can be used for correction.

The second method is to use a color chart. It is a method of photographing a color chart prepared before taking an object and correcting the color of the multi-point image based on the information. This is the method proposed by Ilie and Joshi [4] [5]. Although this method can be used for accurate correction, it is inconvenient to use a color chart. In addition, it has a disadvantage that it can not be applied to an image in which a color chart is not photographed before photographing.

The third method is to find the corresponding pixel value in each image, compare the color information of that part, and then correct the whole image based on the value. [Reference 6] [Document 7]. In order to find correspondences between images obtained at different points in time, a block matching method or a matching method based on a feature point may be used. Performance differences may occur depending on the feature point extraction technique.

ITU-T RECOMMENDATION H.264 "Advanced Video Coding for Generic Audiovisual Services," May, 2003. [Document 2] C. Doutre, P. Nasiopoulos, "A Color Correction Preprocessing Method for Multiview Video Coding," Department of Electronic and Computer Engineering, University of British Columbia. [3] U. Fecker, M. Barkowsky, and A. Kaup, "Histogram-Based Prefiltering for Luminance and Chrominance Compensation of Multiview Video," IEEE Trans. , vol. 18, no. 9, Sep. 2008. [4] G. Jiang, F. Shao, M. Yu, K. Chen, and X. Chen, "New Color Correction Approach to Multi-view Images with Region Correspondence," Lecture Notes in Computer Science, Vol. 4113, Aug. 2006, pp. 1224-1228. [5] K. Yamamoto, M. Kitahara, H. Kimata, T. Yendo, T. Fujii, M. Tanimoto, S. Shimizu, K. Kamikura, and Y. Yashima, "Multiview Video Coding Using Interpolation and Color Correction "'IEEE Transactions on Circuits and Systems for Video Technology, Vol. 17, No. 11, Nov. 2007, pp. 1436-1449. [6] A. Ilie and G. Welch, "Ensuring color consistency across multiple cameras," IEEE International Conference on Computer Vision, Oct. 2005, pp. II: 1268-1275. [7] N. Joshi, B. Wilburn, V. Vaish, M. Levoy, and M. Horowitz, "Automatic color calibration for large camera arrays," in UCSD CSE Tech. Rep. CS2005-0821, May 2005. [Literature 8] A. Bjdrck, Numerical "Methods for Least Squares Problem, SIAM," Philadelpia, 1996. [Document 9] C. Harris and M.J. Stephens. A combined corner and edge detector. In Alvey Vision Conference, pages 147-142, 1988. [Literature 10] Hyo-Sung Kim, Sung-Yeol Kim, "3DTV Three Dimensional Information Processing", pp. 116

SUMMARY OF THE INVENTION The object of the present invention is to solve the problems described above. In consideration of the inaccuracy of the first method and the inconvenience of the second method, the third method, the feature point-based color correction method, Color compensation method to provide a new preprocessing method.

In particular, it is an object of the present invention to provide a method and apparatus for modeling a relative camera characteristic based on corresponding minutiae points between multi-view images in order to correct a color difference between adjacent points due to differences in characteristics between illumination and cameras, To provide a preprocessing method of multi-view video.

More specifically, the present invention uses the Harris corner detection method to extract the corresponding feature points, estimates the coefficients of the modeled formula by the Gauss-Newton circulation algorithm, And to provide a preprocessing method of multi-viewpoint video based on feature points for correcting the color values of the target video by RGB components.

According to an aspect of the present invention, there is provided a preprocessing method of multi-view video based on feature points for color component correction, which receives and corrects at least two multi-view images having different viewpoints, (Hereinafter referred to as a camera characteristic model) for estimating an image (hereinafter referred to as a correction view image) from another multi-view image (hereinafter referred to as a reference view image); (b) extracting feature points corresponding to each other in the corrected viewpoint image and the reference viewpoint image; (c) estimating the corresponding point by applying the corresponding point to the characteristic model with respect to the minutiae point of the reference point image (hereinafter referred to as a reference point) and the minutiae point of the correction-point-of-view image corresponding to the reference point (Hereinafter referred to as an estimation point) and a coefficient of the characteristic model (hereinafter referred to as a camera characteristic coefficient) that minimizes the difference between the reference point and the reference point; And (d) correcting the corrected viewpoint image with the feature coefficient.

Further, the present invention is characterized in that, in the preprocessing method of multi-view video based on feature points, the camera characteristic model uses [Expression 1].

[Equation 1]

Where x and y are the pixel value of the corresponding point and the reference point,

a, b, and γ are coefficients.

According to another aspect of the present invention, there is provided a preprocessing method for multi-view video based on feature points, the feature points corresponding to each other at the time of the correction view image and the reference image are extracted using a Harris corner detection algorithm.

Further, the present invention is characterized in that in the preprocessing method of multi-view video based on feature points, a Gauss-Newton circulation algorithm is applied to minimize a difference between the estimated point and the reference point.

를 최소화하는

인 특성 계수 (a,b,γ)를 구하는 것을 특징으로 한다.(X ₁ , y ₁ ), (x ₂ , y ₂ ), ..., (x _m , y) of the m corresponding points and reference points _m )} by applying the Gauss-Newton circulation algorithm,

To minimize

(A, b, and gamma), which are characteristic values of the characteristic coefficients.

[Equation 2]

According to another aspect of the present invention, there is provided a preprocessing method for multi-view video based on feature points, wherein the image is an RGB image, and the coefficients are obtained for each RGB component.

The present invention also relates to a computer-readable recording medium on which a program for performing a preprocessing method of multi-view video based on the feature point is recorded.

As described above, according to the preprocessing method of multi-view video based on the feature point according to the present invention, it is possible to improve the picture quality (PSNR) and the bit rate when encoding a multi-view video image. In particular, the PSNR can be improved by 0.5 dB to 0.8 dB and the bit rate can be reduced by about 14% in comparison with the conventional cumulative histogram-based preprocessing method.

1 is a diagram showing a configuration of an overall system for carrying out the present invention.
2 is an example of illumination mismatch between adjacent viewpoint images.
3 is a flowchart illustrating a preprocessing method of multi-view video based on feature points according to an embodiment of the present invention.
Figure 4 is an example of a graph of relative feature differences of the camera according to the present invention.
FIG. 5 illustrates a process of modeling a relative characteristic of a camera according to an exemplary embodiment of the present invention.
FIG. 6 is an example of extracting mutually corresponding minutiae according to the present invention.
FIG. 7 is a table showing performance in comparison with the prior art according to the experiment of the present invention.
8 is a table showing camera characteristic coefficients extracted according to the experiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the drawings.

In the description of the present invention, the same parts are denoted by the same reference numerals, and repetitive description thereof will be omitted.

First, examples of the configuration of the entire system for carrying out the present invention will be described with reference to Fig.

1, a preprocessing method for multi-view video based on feature points according to the present invention includes receiving multi-view images 60 captured by at least two multi-viewpoint cameras 20 having different viewpoints, May be implemented as a program system on the computer terminal 30 to be corrected. That is, the preprocessing method of the feature point-based multi-view video may be implemented by a program and installed in the computer terminal 30 and executed. The program installed in the computer terminal 30 can operate as one program system 40. [

Meanwhile, as another embodiment, the preprocessing method of multi-viewpoint video based on feature points may be implemented by a single electronic circuit such as an ASIC (on-demand semiconductor) in addition to a general purpose computer which is constituted by a program. Or a dedicated computer terminal 20 that exclusively processes only the correction of the color component with respect to the multi-view image. This is referred to as a preprocessing device 40 for multi-view video. Other possible forms may also be practiced.

The multi-viewpoint camera 20 refers to a camera for photographing a plurality of scenes and objects 10 by arranging at least two cameras in a matrix or matrix.

The photographed image 60 is an image photographed by the multi-viewpoint camera 20 or a multi-view image. The multi-view image 60 is directly input to and stored in the computer terminal 30, and is processed by the preprocessing device 40. [ Alternatively, the multi-view image 60 may be stored in advance in the storage medium of the computer terminal 30, and may be read by inputting the multi-view image 60 stored by the preprocessing device 40.

The multi-view image 60 is an image photographed by each camera 20 and having different viewpoints according to the viewpoints of the respective cameras 20. The multi-view image (60) consists of image frames that are consecutive in time. Also, the image 60 may have one frame (or image). That is, the image 60 corresponds to one image. At this time, each multi-view image 60 corresponds to each frame image of the same time frame.

The preprocessing device 40 obtains the characteristic coefficients of the camera characteristic model using the images of the respective multi-view images 60 corresponding to each other as described above, and corrects the color components for each of the images. The color correction for the multi-view image means correcting each image or frame image, but the term "image" is used below unless there is a need for a particular distinction.

Next, before describing the present invention, the point-to-point illumination and color mismatch phenomenon to be solved in the present invention will be described in more detail.

The inter-view prediction coding of multi-view video is a method of removing spatial redundancy between adjacent view images and compressing them. It is a very effective coding technique with inter-frame motion prediction method in multi-view images, but it may cause inaccurate prediction due to illumination and color discrepancy between adjacent points. This may result in lower coding efficiency of multi-view video.

Figs. 1 (a) and 1 (b) show examples of illumination and color mismatch between neighboring viewpoint images in the <Uli> and <Ballroom> experimental images of the multi-view video, respectively. As shown in FIG. 1, since the multi-view video images are obtained from cameras existing at different positions, illumination and color discrepancy may exist between different view images.

In the < Uli > image of Fig. 1 (a), the skin color of the character in the viewpoint image of the fourth viewpoint is red compared to the viewpoint image of the third viewpoint obtained by the camera No. 3, The color of the curtain and the stairs in the time-point image is darker than that of the time point image obtained in the 5th camera.

The characteristic of the multi-view video is a cause of erroneous prediction in the inter-view prediction referring to the adjacent view image, resulting in a decrease in the encoding performance. Such discrepancies may exist between the front and back frames of a single view video due to differences in illumination over time. However, in general, illumination and color mismatches in a single viewpoint image are less influenced by the different viewpoints of the different cameras than those of other images, and thus the effect on coding efficiency is relatively small.

There are two main causes of illumination and color mismatch in multi-view images. The first is that differences in lighting and color values may be caused by problems such as differences in production between cameras that acquire multi-view video and variations in iris control. That is, the internal characteristics of the camera are different. Another cause is the location variation of different cameras. Depending on the location of the camera, the ambient lighting can be affected differently and the result can be reflected in the image. Even if the cameras have the same internal characteristics, if they have different relative positions and directions, they will have a local effect on the image. These factors lead to illumination and color discrepancies in the image. In fact, it is easy to see that there are many differences between images when you acquire the same objects with different cameras at different viewpoints.

Next, a preprocessing method of multi-viewpoint video based on feature points according to an embodiment of the present invention will be described with reference to FIG. 3 to FIG.

As shown in FIG. 3, the method according to an exemplary embodiment of the present invention includes a camera characteristic model setting step S10, a feature point extracting step S20, a feature coefficient obtaining step S30, (Step S40).

That is, a camera characteristic model for color correction is set (S10). The camera characteristic model is a relational expression for estimating one multi-view image (hereinafter referred to as a correction view image) from another multi-view image (hereinafter referred to as a reference view image). Then, feature points are extracted to search corresponding points in each viewpoint image. In the present invention, a Harris corner detection method is used (S20). The relative characteristics of each camera are modeled and coefficient values in the camera characteristic model are estimated through a Gauss-Newton circulation algorithm based on the extracted feature points (S30). The estimated coefficients are applied to the camera characteristic model to correct the color of the image (S40).

First, a camera characteristic model setting step (S10) according to an embodiment of the present invention will be described in more detail.

The method according to the present invention utilizes the correspondence point-based color correction method as a preprocessing method for multi-view video coding considering the disadvantages of the histogram-based color correction method and the color chart-based correction methods mentioned in the background art.

When acquiring the multi - view image, various characteristics of the camera may affect the color of the acquired image. In addition, even if the camera of the same model is used and the camera setting is the same, the characteristic inherent to each camera can not be made the same. Therefore, in order to acquire multi-view video, post-processing of artificial color correction is required even if the same camera is used. This means that the relative characteristics of each camera should be modeled as an equation to control the factors affecting the characteristics of the camera.

Relative camera characteristics mean the relative difference in how much the camera perceives light when light of the same intensity is incident on two cameras. The main camera characteristics that cause the difference are gain, off-set, and gamma values [9].

FIG. 4 illustrates differences in color values between a reference view camera and a corresponding point-of-view camera that needs to be corrected by modeling each characteristic difference. The reference view image is the reference image for color correction and the target view image is the image to be corrected according to the reference view image.

4, y _ref denotes a pixel value of the reference view image, and x _tar denotes a pixel value of the correction view image. a, b, and gamma are coefficients for gain, offset, and gamma, respectively. After each coefficient is formulated, the final modeling considering the characteristics of the coefficients simultaneously is shown in FIG.

This model defines the characteristics difference between the reference point camera and the correction point camera and derives the characteristic coefficients of the modeled equation and corrects the color based on this.

Meanwhile, since the method according to the present invention receives and corrects at least two multi-view images having different viewpoints, a camera characteristic model should be obtained for each of two images among a plurality of multi-view images. That is, a camera characteristic model, which is a relational expression for estimating one multi-view image (or a correction view image) from another multi-view image (or reference view image) is set.

At this time, the camera characteristic model for the correction time point and the reference time point image can be expressed as the following Equation (1).

[Equation 1]

However, x and y are the pixel values of the reference time point and the correction time point image corresponding to each other. That is, y is a feature point (or reference point) of the reference view image, and x is a feature point (or corresponding point) of the correction point corresponding to y. a, b, and gamma are coefficients as described above and correspond to the camera characteristic coefficient of the camera characteristic model.

In the following steps, corresponding feature points of the correction time point and the reference point image are extracted and the camera characteristic coefficients a, b, and gamma are estimated using the reference points and the corresponding points. Then, the estimated camera characteristic coefficient is substituted into the above equation (1), and the color of the image is corrected.

Next, step S20 of extracting feature points according to an embodiment of the present invention will be described in more detail.

In order to estimate the above-mentioned characteristic coefficients, it is necessary to acquire corresponding color information from the reference view image and the correction view image. That is, feature points corresponding to each other are extracted from the correction point and the reference point image. In the present invention, the corresponding points of two view points are extracted using the Harris corner detection algorithm, and coefficients are estimated based on the information [10]. The Harris corner detection method analyzes the change in pixel values in a matching window (matching window), and finds a point having a largest change in pixel value among the values over a threshold value for a predetermined upper, lower, left, and right sides as a corner point.

The Harris corner detection technique basically has a window that moves vertically and horizontally in the image, and a method of finding the corner by analyzing the change of pixel values in the window. It has the advantage that it is not affected by image size change or rotation. Therefore, since only the information of the corner feature is extracted regardless of the occlusion area or non-occlusion area which may exist in the multi-view image, the color information of the corresponding point can be extracted more accurately.

FIG. 6 shows the result of extracting correspondence points from the <Breakdancers> image through the Harris corner method. Corresponding points extracted from two images at different points in time (or between the correction point and the reference point image) are shown by connecting with a dotted line. Since the positions of the corner points in the image are the same in the two images, several corner points are found and used for the estimation of the characteristic coefficient.

Next, the step S30 of estimating the characteristic coefficient according to the embodiment of the present invention will be described in more detail.

The sample values extracted in the previous step are stored in the memory buffer for each of the RGB channels. This is to estimate the coefficient of property by RGB component. On the other hand, the pixel values of the feature points corresponding to each other in the two images (or the reference point and the correction point image) are sample values. That is, the sample value is a pixel value of a minutiae point (or a reference point) of a reference point image and a minutiae point (or a corresponding point) of a correction point image corresponding to the reference point, and a pair of reference points and corresponding points corresponds to one sample.

Based on the sample value, a characteristic coefficient is estimated. Relative camera models between two cameras in different locations are nonlinear, so it is difficult to obtain them directly. Therefore, each coefficient is estimated through repetitive calculation of the Gauss-Newton circulation algorithm [Document 8].

The Gauss-Newton method is a generalized numerical method used to obtain a minimum square estimate of the error. This has the advantage of reducing unnecessary calculations by taking advantage of the characteristics of the second derivative. The form of the sum of the square shape functions for the m sample pixel values is as shown in the following equation (2).

&Quot; (2) &quot;

The Gauss-Newton method finds the minimum value using the cyclic method from the initial predicted value. r is an error value obtained by subtracting the function value of the variable predicted from the actual value, and is given by Equation (3).

&Quot; (3) &quot;

는 특성 계수 {a,b,γ}로 이루어진 벡터이다. 따라서 함수 f는 다음과 같이 나타낼 수 있다.Here, y _i denotes a sample pixel value of the reference view image existing in the memory buffer, and x _i denotes a pixel value of the corresponding correction view image. The function f means an equation modeled in Equation (1).

Is a vector composed of characteristic coefficients {a, b, y}. Therefore, the function f can be expressed as follows.

&Quot; (4) &quot;

가 최소값을 갖게 될 때까지 반복하여 카메라의 특성 계수를 추정하게 된다. 메모리에 샘플값의 RGB 성분을 분리하여 저장하였으므로 세 가지 특성 계수 {a,b,γ}가 RGB 성분별로 각각 세 개가 존재하며 결국 카메라 당 총 아홉 개의 계수가 구해진다.
Nonlinear function that means square sum of error

Is repeated until the minimum value is obtained. Since the RGB components of the sample values are stored in the memory, there are three each of the three characteristic coefficients {a, b, γ} for each of the RGB components, and finally a total of nine coefficients per camera are obtained.

Next, effects of the present invention will be described in more detail with reference to Figs. 7 to 8. Fig.

In the experiment according to the present invention, an experiment is performed to verify the performance of the proposed technique using <Ballroom>, <Breakdancers>, <Uli> and <Average> Respectively. First, Fecker's accumulated histogram-based preprocessing method [2] and the new preprocessing method proposed by the present invention are applied to multi-view video coding.

As shown in FIG. 7, the difference from the case in which the preprocessing is not performed at all is compared in terms of the picture quality (PSNR) and the compression rate (bit rate). In the picture quality according to the present invention, 0.5dB and showed similar results for Cb. In terms of compression efficiency, the method according to the present invention shows an average improvement of about 14%.

FIG. 8 is a characteristic coefficient of a relative camera model calculated through the Gauss-Newton circulation technique. In the <Uli> experimental image, the coefficient of characteristic of the 4th camera (point 4) was estimated. These coefficients are used for color correction at two points.

In the present invention, a correspondence point based correction method using feature points is proposed as a preprocessing method to solve illumination and color discrepancy problems that may occur in multi-view video coding. The feature points of the image are extracted and the coefficient of property is estimated by numerical method based on the correspondence relation of various pixel values. Feature point extraction was performed using Harris corner detection method. When the proposed method is applied to multi - view video coding, it can be confirmed that the coding performance is improved.

As mentioned above, although the invention made by this inventor was demonstrated concretely according to the said Example, this invention is not limited to the said Example, Of course, a various change is possible in the range which does not deviate from the summary.

10: Object 20: Multi-view camera
30: computer terminal 40: program system
60: Video

Claims (7)

A method for preprocessing a multi-viewpoint video based on a feature point for color component correction, the method comprising the steps of: receiving at least two multi-
(a) setting a relational expression (hereinafter referred to as a camera characteristic model) for estimating one multi-view image (hereinafter referred to as a correction view image) from another multi-view image (hereinafter referred to as a reference view image);
(b) extracting feature points corresponding to each other in the corrected viewpoint image and the reference viewpoint image;
(c) estimating the corresponding point by applying the corresponding point to the characteristic model with respect to the minutiae point of the reference point image (hereinafter referred to as a reference point) and the minutiae point of the correction-point-of-view image corresponding to the reference point (Hereinafter referred to as an estimation point) and a coefficient of the characteristic model (hereinafter referred to as a camera characteristic coefficient) that minimizes the difference between the reference point and the reference point; And
and (d) correcting the corrected view point image using the feature coefficients.
The method of claim 1,
The camera characteristic model uses [Equation 1], characterized in that the feature-based multi-view video pre-processing method.
[Equation 1]

Where x and y are the pixel value of the corresponding point and the reference point,
a, b, and γ are coefficients.
The method of claim 1,
And extracting feature points corresponding to each other at a time point of the corrected viewpoint image and a reference point of view using a Harris corner detection algorithm.
The method of claim 1,
And a Gauss-Newton cyclic algorithm for minimizing the difference between the estimated point and the reference point.
5. The method of claim 4,
For a pair of m corresponding points and reference points {(x ₁ , y ₁ ), (x ₂ , y ₂ ), ..., (x _m , y _m )}, the Gauss-Newton cyclic algorithm is applied to Equation 2]

To minimize

(A, b,?) Of the feature point based on the feature points.
[Equation 2]

5. The method according to any one of claims 1 to 4,
Wherein the image is an RGB image and the coefficients are obtained for each of the RGB components.
A computer-readable recording medium having recorded thereon a program for performing the method for preprocessing the feature-based multi-view video according to any one of claims 1 to 5.

Images