KR101285810B1 - A Super-Resolution method by Motion Estimation with Sub-pixel Accuracy using 6-Tap FIR Filter - Google Patents

Abstract

The present invention relates to a method for reconstructing a subpixel-based motion estimation-based superresolution image which receives continuous low-resolution images and reconstructs a high-resolution image by subpixel motion estimation using a 6-Tap FIR filter. Receiving a low resolution image, wherein the low resolution image comprises a target image and a time image before and after the target image; (b) obtaining a subpixel by applying interpolation to the low resolution images, and obtaining an image (hereinafter referred to as a subpixel image) including pixels and subpixels of the low resolution images; (c) obtaining a motion vector by searching for an optimal block match for each of the subpixel images (hereinafter referred to as the target subpixel images) and the before and after time images (hereinafter, referred to as subpixel images) of the target image; And (d) finding a pixel (hereinafter, a matching point) of the front and rear subpixel image matched to the target point by a motion vector, for each pixel of the target subpixel image (hereinafter, the target point), and interpolating the target point to the matching point. Provide a configuration to include.
By using the super resolution image reconstruction method as described above, more accurate high frequency components can be reconstructed from a continuous low resolution image through motion estimation of subpixel precision.

Description

Super-Resolution method by Motion Estimation with Sub-pixel Accuracy using 6-Tap FIR Filter}

According to the present invention, a subpixel motion estimation-based superresolution image reconstruction that receives continuous low resolution images (frames) and reconstructs a high resolution image through subpixel motion estimation using a 6-tap finite impulse response (FIR) filter It is about a method.

In general, a technique of inserting an appropriate value between existing pixels to enlarge an image is called image interpolation. Recently, with the development of digital image acquisition media such as digital cameras, the importance of high resolution image interpolation is increasing in various fields. In the medical field, for example, high-resolution medical images are very important for doctors. This is because the performance of computer pattern recognition, which is frequently used in this field, is highly dependent on the resolution of the input image.

Image interpolation is also closely related to de-interlacing, which converts interlaced images into sequential scans, and super-resolution techniques to reconstruct high-resolution images from multiple low-resolution images. have. An image obtained by the digital image medium may be modeled as a down sampled image after the high resolution image passes through the low frequency band pass filter. An image obtained through this process loses information of many high frequency components due to an aliasing phenomenon. Therefore, effective reconstruction of such high frequency components is the most important technique in image interpolation.

Traditional image interpolation is based on the weighted-sum of low resolution image pixels, and the representative methods are the nearest neighborhood interpolation, bi-linear interpolation, and bi-cubic. Interpolation, etc. These methods can be interpreted as the concept of a linear filter. That is, the low resolution image becomes an input signal of the filter, and the weights multiplied by the image become filter coefficients. The process of enlarging the entire image may be regarded as applying the filter after upsampling the low resolution input image. Therefore, the performance of the interpolation method can be analyzed by confirming the frequency characteristics of the weights that become the filter coefficients. From this point of view, most of the weight-based methods have low-pass filter characteristics and thus cannot effectively recover high frequency components damaged by aliasing.

To solve this problem, a plurality of consecutive frames can be used to restore high frequency components with higher accuracy than using a single image.

SUMMARY OF THE INVENTION An object of the present invention is to solve the problems described above, and receives a low-resolution image (frame) of continuous time, and performs a high-resolution image through sub-pixel unit motion estimation using a 6-Tap finite impulse response (FIR) filter. The present invention provides a method of reconstructing a sub-resolution image based on subpixel unit motion estimation.

That is, an object of the present invention is to find a high-frequency component in a target image in a plurality of frames before and after the target image in sub-pixel unity motions, and then use the sub-pixel unit to interpolate the high-frequency component. It is to provide a method for motion estimation based super resolution image reconstruction.

In this case, since more accurate motion estimation determines the quality of the resultant image, motion estimation with sub-pixel precision is required, not an integer unit. For this purpose, 6-tap of H.264 / AVC is required to obtain sub-pixels of input images. Use an FIR filter. After that, after sub-pixel precision motion estimation is applied to the target image, the empty space of the target image is interpolated using the two-dimensional interpolation method proposed by D. Shepard.

In order to achieve the above object, the present invention relates to a sub-pixel unit motion estimation-based super-resolution image reconstruction method for receiving continuous low-resolution images and reconstructing a high-resolution image through sub-pixel unit motion estimation. Receiving a low resolution image having a normal time, wherein the low resolution image comprises a target image and a time image before and after the target image; (b) obtaining a subpixel by applying interpolation to the low resolution images, and obtaining an image including a pixel of the low resolution images and the subpixel (hereinafter referred to as a subpixel image); (c) obtaining a motion vector by searching an optimal block match for each of the subpixel images (hereinafter, referred to as target subpixel images) and the before and after time images (hereinafter, referred to as subpixel images) of the target image; And (d) for each pixel (hereinafter, referred to as a target point) of the target subpixel image, to find a pixel (hereinafter referred to as a matching point) of the front and rear subpixel image matched to the target point by the motion vector, and matching the target point to the match. And interpolating to points.

The present invention also provides a method for reconstructing a subpixel-based motion estimation-based super resolution image, wherein the before and after subpixel image is composed of two consecutive previous time images and two subsequent time images of the target image.

In addition, according to the present invention, in the subpixel unit motion estimation-based super resolution image reconstruction method, in step (b), one-half subpixels of the subpixels are applied with a 6-tap FIR filter. The quarter subpixel is obtained by applying a bi-linear interpolation method to the half subpixel.

In addition, according to the present invention, in the subpixel unit motion estimation based super resolution image reconstruction method, in step (c), the optimal block matching is performed in units of blocks between the target subpixel image and the front and rear subpixel images. of absolute difference) operation to select a block having the minimum SAD as an optimal block.

In addition, the present invention is a subpixel unit motion estimation-based super-resolution image reconstruction method, characterized in that the block unit is 2 * 2.

In addition, the present invention is characterized in that in the sub-pixel unit motion estimation-based super-resolution image reconstruction method, the matching block is searched by a global search method or a spiral search method to select the optimal block.

The present invention also provides a subpixel unit motion estimation based super resolution image reconstruction method, wherein in step (d), the target point interpolates the matching points by a weight based on a distance between the target point and the matching point. It is done.

In another aspect of the present invention, in the subpixel unit motion estimation based super resolution image reconstruction method, in step (d), the target point is interpolated between the matching points by Equation 1.

[Equation 1]

The present invention also relates to a computer-readable recording medium having recorded thereon a program for performing the method of performing subpixel unit motion estimation based super resolution image reconstruction.

As described above, according to the subpixel unit motion estimation based super resolution image reconstruction method according to the present invention, an effect of reconstructing a high frequency component more accurately from a continuous low resolution image through motion estimation with subpixel precision is obtained.

In particular, the experimental results show that subjective and objective results are slightly superior to images obtained using the existing nearest neighborhood interpolation, bi-linear interpolation, and bi-cubic interpolation.

1 is a diagram showing a configuration of an overall system for carrying out the present invention.
2 is a flowchart illustrating a method of reconstructing a super resolution image based on subpixel unit motion estimation according to an embodiment of the present invention.
3 is a diagram illustrating a process of making a low resolution image according to an embodiment of the present invention.
Figure 4 illustrates an example of aliasing that occurs during the ADC process according to the present invention.
Figure 4 illustrates an example of aliasing that occurs during the ADC process according to the present invention.
5 illustrates an example in which a pixel moves in an integer unit or a subpixel unit according to the present invention.
FIG. 6 shows an example of applying a 6-tap FIR filter during 1/2 pixel interpolation according to the present invention.
7 illustrates an example of a global search method according to the present invention.
8 shows an example of a spiral search method according to the present invention.
9 shows an example of the positional relationship between the image and the position of the interpolation target point according to the motion estimation according to the present invention.
10 shows an example of the PSNR measurement results when four, eight, and twelve data points according to the present invention are used.
11 shows an example before and after removing the grid phenomenon of the image outline according to the present invention.
12 to 15 show the original image, the shortest-entry, double-linear, and higher for "Foreman (9th frame)", "Car (8th frame)", "Mother-daughter (6th frame)" according to the present invention. An example of an order and an image to which the high resolution method of the present invention is applied are shown.

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.

As shown in FIG. 1, the subpixel unit motion estimation based super resolution image reconstruction method according to the present invention receives a low resolution image (or frame) 10 and performs motion estimation of a subpixel unit with respect to the frame (or image). And a program system on the computer terminal 20 to restore the high resolution image. That is, the super resolution image restoration method may be configured as a program and installed and executed in the computer terminal 20. The program installed in the computer terminal 20 may operate like one program system 30.

On the other hand, as another embodiment, the super-resolution image restoration method may be implemented as a program and operate in a general-purpose computer, and may be implemented as one electronic circuit such as an ASIC (custom-type semiconductor). Alternatively, the present invention may be developed as a dedicated computer terminal 20 which processes only a low resolution image composed of continuous frames to be restored to a high resolution image. This is called a super resolution image reconstruction device. Other possible forms may also be practiced.

The low resolution image 10 is an image composed of a plurality of frames in continuous time. For example, if the current frame is a frame at time t, the previous frame is a frame at times t-1, t-2, t-3, and the like, and the frames after t + 1, t + 2, t + 3. , ..., and so on at time. These frames will be referred to as front and rear frames. An image consists of a number of frames that are contiguous in time. That is, the image is composed of frames of time zones ..., t-3, t-2, t-1, t, t + 1, t + 2, t + 3, .... In this case, the current frame, that is, the frame of interest is referred to as a target frame. In addition, although the image is composed of a plurality of frames, in the following, unless there is a need for distinction, the frame and the image are mixed.

Next, a method of reconstructing a sub-resolution image based on subpixel unit motion estimation according to an embodiment of the present invention will be described with reference to FIG. 2.

As shown in Figure 2, the method of extracting the rotational invariant texture feature according to an embodiment of the present invention (a) receiving a series of low-resolution images in time (S10); (b) obtaining a subpixel image including a subpixel by applying interpolation to low resolution images (S20); (c) obtaining a motion vector by searching an optimal block match with respect to the subpixel image and the before and after time image of the target image (S30); And (d) viewing the target point as a matching point of the before and after subpixel image matched with the target point of the target subpixel image (S40).

More specifically, in an embodiment of the present invention, first, after each of two frames (hereinafter, before and after frames, before and after images based on the image (or target image and target frame) to which the super resolution method is to be applied A total of five low resolution images, including), is input (S10). In order to obtain sub-pixels on the five images, a 6-tap finite impulse response (FIR) filter is applied (S20). Afterwards, the high frequency portion of the target image is found in another low-resolution image by the motion estimation in the unit of 1/4 pixel (S30), and the resolution of the target image is increased by using these points (S40).

Before specifically describing a method of restoring a super resolution image according to an embodiment of the present invention, the principle of restoring the super resolution image will be described in more detail with reference to FIGS. 3 to 5.

In the present invention, it is assumed that the low resolution image has an observation model as shown in [Equation 1].

[Equation 1]

Where y _k is the k-th low resolution image, D is the downsampling matrix, and x _k and n _k are the k-th high resolution image and the noise added to the corresponding image, respectively. In the present invention, the process of restoring the target image x _k is shown by the inverse of the observation model of [Equation 1] shown in FIG. 3 and the result.

The image that the human eye sees is a continuous analog signal. In order to convert a video signal into a digital signal that can be processed by a computer, a process of digitizing the signal such as a voice signal is required. In this process, aliasing due to the sampling interval occurs in the process of becoming a low resolution image due to the performance limitation of a digital device such as an image sensor as shown in FIG. 4. As a result, high frequency components of the original video signal are lost.

Assuming that the current frame in the video is f (t) and the previous frame is f (t-1), the two images will have only a slight difference when they are not transition parts of the scene. 5 illustrates the movement of an object existing between frames. In FIG. 5, the left column represents an f (t-1) frame and the right column represents an f (t) frame.

5 (a) shows a case in which the movement of an object between frames moves by an integer unit which is a sampling interval. In this case, the digitized image has the same outline information of the corresponding object of f (t-1) and f (t). In the case of FIG. 5B, an object between frames moves in a sub-pixel unit smaller than the sampling interval. In this case, unlike the first case, the two images have different contour information about the same object.

If multiple images have the same high frequency components as in the first case, super-resolution techniques cannot be applied to the images. However, in the second case, super-resolution technique can be applied by collecting different high frequency information. Therefore, in the present invention, the high frequency component is restored by using the second case moved to the subpixel unit.

Next, a method of reconstructing a super resolution image according to an embodiment of the present invention will be described in detail. Hereinafter, a low resolution image of a continuous time is input, and the received low resolution image is restored to a super resolution image. In this case, the low resolution image is composed of a target image and a time image before and after the target image. In particular, in the following embodiments, a total of five low-resolution images are input, including two frames (before and after time images) before and after the images (target images) to which the method of the present invention is to be applied. . That is, the before and after subpixel image is composed of two consecutive previous time images and two subsequent time images of the target image.

First, a subpixel is obtained by applying interpolation to low resolution images, and an image (hereinafter referred to as a subpixel image) including a pixel of the low resolution images and the subpixel is obtained (S20). In particular, one-half subpixel of the subpixel applies a six-tap finite impulse response (FIR) filter, and one-quarter subpixel applies bi-linear interpolation to the half subpixel. Obtain by applying

That is, in order to obtain the above-described sub-pixel motion information, the sub-pixels of the input images must first be found. In an embodiment of the present invention, a sub-pixel is found by applying a 6-tap finite impulse response (FIR) filter of H.264 / AVC of Equation 2 for each input image.

 &Quot; (2) "

Where b and h each represent pixel values in units of 1/2 pixel, and A, C, G, M, R, T, E, F, H, I, and J represent original pixel values in integer units, respectively. . The number multiplied by the pixel of each integer unit represents a weight.

It will be described with reference to Figure 6 to help understand the equation (2). First, the original pixels in the horizontal and vertical directions are weighted to find b and h. A quarter pixel is obtained by obtaining a half pixel and then applying bi-linear interpolation.

As described above, subpixels such as 1/4 pixels and 1/2 pixels are obtained, and an image (or frame) composed of the subpixels and the pixels of the original image is called a subpixel image. Subpixel images are obtained for each low resolution image. That is, since five low resolution images are used in the above example, five subpixel images are also obtained. In this case, the subpixel image of the target image is called a target subpixel image, and the subpixel image of the before and after time image is called a before and after subpixel image.

Next, an optimal block match is searched for each of the subpixel images (hereinafter referred to as the target subpixel images) and the before and after time images (hereinafter, referred to as subpixel images) of the target image to obtain a motion vector (S30). In particular, the optimal block matching performs a sum of absolute difference (SAD) operation in units of blocks between the target subpixel image and the front and rear subpixel images, and selects the block having the smallest SAD as an optimal block.

 After both 1/2 and 1/4 pixels are obtained, motion estimation is performed in units of 2 * 2 blocks. First, an optimal matching block is selected through a sum of absolute difference (SAD) operation. 7 and 8 illustrate a method of searching for different optimal blocks, respectively. FIG. 7 illustrates a full search method. If a 64 * 64 search block is used, 64 * 64 SAD values are required. Equation 3 is used to calculate a single SAD value.

 &Quot; (3) "

Where SAD (i, j) is the SAD value at position (i, j). In an embodiment of the present invention, since 64 * 64 search blocks are used, i and j have values of 0 to 63, respectively. B _t and B _t-1 denote macroblocks of size 2 * 2 of the frame at time t and t-1, respectively. Therefore, x and y have values of 0-1. In particular, B _{t -1} is a macroblock at position (i, j).

Where i and j is a square having a size of 64 * 64 as a center of the search region (search range), the search region (search range) of B _t. In Equation 3, B _t is a 'fixed' 2 * 2 block in the reference image, and B _{t -1} is one pixel unit from (0, 0) to (63, 63) in the search range. It is a 2 * 2 macroblock that moves and yields 64 * 64 SAD values. Equation 3 is based on the assumption that B _t is fixed. As a result, i and j are search ranges, x and y are moving blocks and coordinates in fixed blocks (where x and y both have values of 0 to 1).

In an embodiment of the present invention, since the previous two frames and the subsequent two frames are used, B _t-2 , B _t-1 , B _{t + 1} , and B _{t + 2} have respective SAD equations. That is, while the position of B _t of the target image is fixed, the macro block of the other low resolution image is moved in the search block to fill the SAD value. After all the 64 * 64 SAD values are obtained, the minimum value is found and the motion vector is calculated assuming that the image has shifted in that direction.

There are two types of searching methods. Since the global search shown in FIG. 7 is time consuming, a spiral search for searching near a predetermined range around a fixed block is shown in FIG. 8. Use (spiral search).

FIG. 8 (a) shows a change in coordinates when moving with the upper left point of a 2 * 2 block as a reference point when the spiral search method is selected. FIG. 8 (b) shows the unit of the search range in the spiral search, and it is determined that one kick is searched when moving in the horizontal direction and the vertical direction once.

Using the searched block as described above, motion estimation in units of 1/4 pixels is performed on the remaining four low resolution input images based on the target image. Through this process, the positional relationship between the target low resolution image and the remaining low resolution images can be obtained as shown in FIG. 9 (a) through a motion vector.

In FIG. 9 (a), the circular pixels represent the original pixels of the target image, and the square and diamond pixels each represent a relative position based on the motion vectors of the remaining low resolution images. The portion indicated in FIG. 9B represents target points to be interpolated.

In this case, an original pixel (or a point to be interpolated) of the target image is called a target point, and a pixel of a relative position by a motion vector, that is, a pixel of a relative position (or a corresponding position) in the before and after image is called a matching point. The matching point is a position corresponding to the target point position by the motion vector.

Next, for each pixel (hereinafter, referred to as a target point) of the target subpixel image, a pixel (hereinafter referred to as a matching point) of the front and rear subpixel image matched to the target point by the motion vector is found, and the target point is referred to as the registration point. Interpolate (S40). In particular, the target point interpolates the matching points by a weight by the distance between the target point and the matching point.

When the resolution of the target image is interpolated, insufficient high frequency components are compensated for in another adjacent frame. However, as shown in FIG. 9 (a), when only registration with other low resolution images is completed, it cannot be used as a pixel value of the target interpolation point. It is necessary to obtain a pixel value of the target interpolation point with reference to the matched pixel data. First, determine how many of the closest points to select based on the target interpolation point.

As a criterion for calculating the distance between pixels, Euclidian distance is used as shown in [Equation 4].

 &Quot; (4) "

Where (xx _i ) represents horizontal movement and (yy _i ) represents vertical movement.

The number of data used for interpolation of the target point affects the resulting image. If the number is too small, the information is insufficient to interpolate the high frequency components. On the contrary, if the number is too large, the interpolation method may reduce the accuracy of the interpolation method.

There are several methods for interpolating matched pixel values to target points. In this paper, we use the interpolation method proposed by D. Shepard as shown in [Equation 5].

 [Equation 5]

Equation 5 is an interpolation method using weights based on distances from irregularly spread points based on a corresponding position in a two-dimensional space. In Equation 5, f ₁ (P) represents a value of a target interpolation point (or target point), d _i represents a distance between each target interpolation point and a matching point, and z _i represents a value of a corresponding data point. u is the coefficient of the interpolation equation, and the flatness of the resultant image varies according to this value. If the distance from the matching interpolation point among the collected matching points is 0, the corresponding value is substituted into the target interpolation point. N is the number of matching points used for the target point.

 Next, the effects of the present invention will be described in more detail with reference to FIGS. 10 to 17.

 Experimental images used to test the performance of an embodiment of the present invention are '352', 'Foreman', 'Car', 'Mother-daughter' image. The experiment downsamples the original image into a 176 * 144 size image, and then extracts the first 10 frames of each experimental image from the nearest neighborhood, bi-linear, and bi-cubic interpolation. The invention proceeds back to the original resolution and then compares the subjective and objective results.

For objective performance comparison, the results of the PSNR measurement with the original of the results are shown in FIGS. 10 (a), 10 (b) and 10 (c), respectively. The column of the second row of the table of each figure represents the value of the coefficient u in [Equation 5]. From the third row of each table, the sequence numbers of the images used in the experiments are shown, and the results of using the proposed technique, the double linearity, the higher order, and the shortest-entry point interpolation method are shown in order for the rows forming each scene. 10 (a) to 10 (c), it can be seen that the smaller the coefficient u and the higher the number of used data points, the higher the PSNR value. The difference in PSNR between the existing interpolation methods and the proposed method is similar to that of FIGS. 10 (a) to 10 (c), but it can be confirmed that PSNR values are higher than other interpolation methods under all the experimental conditions.

FIG. 11 (a) shows a lattice phenomenon appearing at the outermost part of an image in a red square in a 6-tap finite impulse response (FIR) filter interpolation process. In most images, the region of interest is located in the center or other region rather than the outer region. Therefore, the bilinear interpolation method is applied to the interpolation of the outermost portion of the image to remove the outer lattice phenomenon and shorten the motion estimation time. It is shown in Figure 11 (b).

12 to 17 show results of subjective quality comparison of each image used in the experiment. For each image used in each experiment, one frame with the best PSNR value was selected. 12, 14 and 16 show the entire part of the resultant image. 12, 14, and 16, it is difficult to easily compare the performance difference between the higher order interpolation and the proposed technique.

Therefore, the resultant image is partially enlarged 6 times in FIGS. 13, 15, and 17. The result of using the shortest-point interpolation method is that the original integer pixel value is copied and inserted into the target interpolation point as it is, so it can be seen that the high frequency component is significantly damaged compared with other results. In the case of the bilinear interpolation method, the step phenomena shown in the shortest point interpolation method can be greatly reduced, but the contour of the object is blurred compared to the original, and as a result, a large amount of high frequency components are lost. The result of using higher-order interpolation in each image is relatively less likely to lose the high-frequency components that blur the contour than the result of using the bilinear interpolation method. It can be seen. It can be seen from FIGS. 13, 15, and 17 that the super resolution image reconstruction method of the present invention substantially compensates for the drawbacks of the bilinear interpolation method and the higher order interpolation method.

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: low resolution image 20: computer terminal
30: Program system

Claims (9)

Sub-pixel-based motion estimation-based super-resolution image reconstruction method that receives a continuous time image (hereinafter, referred to as a low resolution image) and reconstructs an image having a higher resolution than the low resolution image (hereinafter referred to as a high resolution image) through subpixel unit motion estimation. In
(a) receiving a low resolution image of a continuous time, wherein the low resolution image comprises a target image and a time image before and after the target image;
(b) obtaining a subpixel by applying interpolation to the low resolution images, and obtaining an image including a pixel of the low resolution images and the subpixel (hereinafter referred to as a subpixel image);
(c) obtaining a motion vector by searching an optimal block match for each of the subpixel images (hereinafter, referred to as target subpixel images) and the before and after time images (hereinafter, referred to as subpixel images) of the target image; And
(d) For each pixel (hereinafter, referred to as a target point) of the target subpixel image, the pixel (hereinafter referred to as a matching point) of the front and rear subpixel image matched to the target point by the motion vector is found, and the target point is referred to as the matching point. Sub-pixel motion estimation based super resolution image reconstruction method comprising the step of interpolating.
The method of claim 1,
The before-and-after subpixel image is composed of two consecutive images of a previous time of the target image and two consecutive images of a subsequent time of the target image.
The method of claim 1,
In the step (b), one-half subpixel of the subpixel applies a 6-tap FIR filter, and one-quarter subpixel is bilinear to the half subpixel. A subpixel-based motion estimation based super resolution image reconstruction method characterized by applying a linear interpolation method.
The method of claim 1,
In the step (c), the optimal block matching performs a sum of absolute difference (SAD) operation in units of blocks between the target subpixel image and the front and rear subpixel images, so that the block having the smallest SAD is an optimal block. Sub-pixel unit motion estimation based super-resolution image reconstruction method characterized in that the selection.
5. The method of claim 4,

The sub-pixel motion estimation based super resolution image reconstruction method of claim 2, wherein the block unit is 2 * 2.
5. The method of claim 4,
The subpixel unit motion estimation based super resolution image reconstruction method according to claim 1, wherein the matching block is searched by a global search method or a spiral search method to select the optimal block.
The method of claim 1,
In the step (d), the target point is a subpixel unit motion estimation based super-resolution image reconstruction method, characterized in that for interpolating the matching points by the weight of the distance between the target point and the matching point.
The method of claim 7, wherein
In the step (d), the target point is a subpixel unit motion estimation based super-resolution image reconstruction method, characterized in that for interpolating the matching points by [Equation 1].
[Equation 1]

Where P is the target point,
f ₁ (P) is the value obtained by interpolating the target point P (hereinafter referred to as the value of the target interpolation point),
f ₁ () is the interpolation function,
D _i is each registration point,
d _i is the distance between the target point P and each registration point D _i ,
z _i is the value of its matching point D _i ,
u is the coefficient of the interpolation equation,
N is the number of matching points used for the target point,
i is the index of the registration point.
A computer-readable recording medium having recorded thereon a program for performing the subpixel unit motion estimation based super resolution image reconstruction method according to any one of claims 1 to 8.

Images