KR101235488B1 - De-interlacing Apparatus Using Weight Based on Wiener Filter and Method of Using the Same - Google Patents

Abstract

Disclosed are an apparatus and a method for performing deinterlacing using pixels located around a target pixel of interpolation in one frame. The candidate pixel value generator and the candidate pixel value generator respectively generate candidate pixel values corresponding to the directionality of the edges in the diagonal directions P and Q and V directions perpendicular to each other. The present invention provides a deinterlacing apparatus for applying interpolation of pixels by applying a weight α calculated by using a Wiener filter and applying a weighting-based weight to the Wiener filter including a pixel interpolation unit for interpolating pixels. Accordingly, the present invention provides a pixel interpolation apparatus and method with improved image quality, and a relatively small amount of calculation, thereby allowing interpolation of high quality images within a fast processing time.

Description

De-interlacing Apparatus Using Weight Based on Wiener Filter and Method of Using the Same}

The present invention relates to an image processing technology, and more particularly, to an apparatus and method for performing deinterlacing using pixels located around a target pixel of interpolation in one frame.

There are two types of screen formats of an image, a progressive scan method and a progressive scan method. In the progressive scan method, one screen (frame) is divided into two fields for transmission, and in order to form one screen (frame), the first field is skipped and the first field is added to the odd scan line. Then, the second field is scanned in the even-numbered scanning line by skipping the odd-numbered line. The parallel scan method has the advantage of doubling the frame rate while using the same bandwidth. However, this method may cause deterioration of image quality such as flickering of images due to the characteristics of the scanning method. Various deinterlacing techniques have been proposed to prevent such deterioration of image quality.

In particular, as the demand for HDTV display devices such as PDPs and LCDs increases, a deinterlacing process for converting a video signal of a progressive scan type into a video signal of a progressive scan type is essential in order to obtain good image quality. In other words, deinterlacing is to convert empty image data (line) in a frame into one intact frame by using surrounding image data due to the feature of a parallel scan method.

The de-interlacing method is largely classified into an intra-field deinterlacing method using only the information of the current field without using temporal information and an inter-field deinterlacing method using temporal information and the information of the current field. can do. In general, the deinterlacing method using temporal information is excellent in terms of performance. However, due to the large amount of computation, a complex hardware structure is required and serious image quality deterioration occurs when accurate motion detection is not performed. In addition, the deinterlacing method using temporal information should basically utilize the information of the current field. Intra-screen de-interlacing method using current field information is approached by using various spatial filters and approached by considering the direction of edge. Among them, the method of considering the direction of edge is widely used. .

Edge-based line average (ELA) is a method of deinterlacing that considers the directionality of edges, and thus, the operation is fast by interpolating pixels by detecting three directionalities. In addition, ELA has been widely used due to the ease of implementation, but since the interpolation of pixels using only the directional correlation of pixels, there is a problem in that image quality is degraded due to noise. Efficient ELA (E-ELA) is an improvement on ELA that detects a more accurate edge direction and interpolates pixels. The E-ELA classifies the spatial form of the image into three in order to find more accurate orientation than the ELA method. That is, the E-ELA determines the directionality of the edge by dividing the spatial direction into the diagonal directions of P and Q and the vertical direction of V. However, although the E-ELA has improved performance to some extent compared to the ELA method, there is a problem in that the image quality deterioration is not sufficiently reduced due to the inadequate determination of the edge direction in the complex high frequency region.

Modified Edge-based Line Average (MELA), which is modified from ELA, is superior in performance because it detects five directions by interpolating pixels by extending the direction than ELA. However, like the ELA, the MELA has a disadvantage in that the direction is miscalculated, which may cause deterioration of image quality.

In addition, the covariance-based adaptive deinterlacing (CAD) technique obtains the characteristics of a low resolution block using the principle of geometric duality on the assumption that the pixel to be interpolated can be obtained by the weighted sum of eight neighboring pixels. This is then applied to the high resolution block. However, CAD technology that does not consider the edge direction has a disadvantage in that the calculation amount is large.

An object of the present invention for solving the above problems is to provide a de-interlacing apparatus that reduces the phenomenon of deterioration of image quality due to noise in interpolating pixels in one frame.

Another object of the present invention for solving the above problems is to provide a de-interlacing method for reducing the phenomenon of deterioration of image quality due to noise in interpolating pixels within one frame.

In order to achieve the above object, the present invention provides candidate pixel values for generating respective candidate pixel values corresponding to the directionality of edges appearing in different diagonal directions P and Q and V directions perpendicular to the pixels to be interpolated. A pixel calculated by using a Wiener filter is applied to each candidate pixel value generated by the generator and the candidate pixel value generator, and the candidate pixel values to which the weight α is applied are weighted to add pixels. The present invention provides a deinterlacing apparatus for interpolating pixels by applying a Wiener filter-based weight including a pixel interpolation unit to interpolate.

Here, the deinterlacing device for applying the Wiener filter based on the weight of the interpolation pixel is to enlarge the window size until the area attribute coefficient (D _i) the calculated area attribute coefficient (D _i) is greater than a preset value, the window The apparatus may further include a window size determiner that determines a size.

Here, the candidate pixel value generation unit detects the directionality of the edges appearing in different diagonal directions of the P and Q directions and the V direction perpendicular to each other , and an adjacent edge placed in a direction based on the directionality of the edges detected by the edge direction detector. n (where n is an integer) is applied to a fixed filter coefficient ( h (k) ) in the fixed directional interpolation filter (FDIF) as the candidate weight (h), and n adjacent neighbors to which the candidate weight (h) is applied And a candidate pixel value extractor configured to weight the pixels to extract candidate pixel values in the P, Q, and V directions.

)을 생성하고, 행렬식

으로부터 가중치(α)를 연산하는 것을 특징으로 할 수 있다. Here, the pixel interpolation unit has candidate pixel values in the P, Q, and V directions for the peripheral pixels located in the top row and the bottom row in the window centered on the pixel X (i, j) to be interpolated , respectively. To extract the matrix of post-pixel values (

), Determinant

It can be characterized by calculating the weight α from.

Another aspect of the present invention for achieving the above object is to generate respective candidate pixel values corresponding to the directionality of the edges appearing in the V direction perpendicular to the different diagonal directions P, Q direction about the pixel to be interpolated The candidate pixel value calculated by using a Wiener filter is applied to each candidate pixel value generated in the candidate pixel value generating step and the candidate pixel value generating step, and the candidate pixel value to which the weight α is applied. Provided is a deinterlacing method for interpolating pixels by applying a Wiener filter-based weighting step including weighted summation to interpolate pixels.

When using a deinterlacing device that interpolates pixels by applying a Wiener filter-based weight as described above, the image quality may be improved by reducing the phenomenon of deterioration of the screen by incorrectly detecting the directionality of an edge.

In addition, the de-interlacing apparatus and method provided by the present invention not only provide a pixel interpolation apparatus and method with improved image quality, but also have a relatively small amount of calculation, and thus have an advantage of interpolating a high quality image within a fast processing time.

The effects of the present invention are not limited to the effects mentioned above, and other effects not mentioned can be clearly understood by those skilled in the art from the description of the claims.

1 shows a window used to determine the edge direction.
2 is a schematic diagram of the direction of the edge according to the M-ELA method.
FIG. 3 is a window showing a pixel column centered on an interpolation pixel to explain a deinterlacing apparatus according to an embodiment of the present invention.
4 illustrates peripheral pixels partitioned by window size WS to explain a deinterlacing apparatus according to an embodiment of the present invention.
5 is a flowchart illustrating a deinterlacing method according to an embodiment of the present invention.
FIG. 6 is a table illustrating objective performance of various deinterlacing schemes in terms of peak signal to noise ratio (PSNR).
FIG. 7 is a table illustrating evaluation of subjective picture quality of various deinterlacing schemes using MSSIM.
8 is a photograph showing a result of deinterlacing an image according to an embodiment of the present invention and a result of deinterlacing a conventional method.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail. It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. Like reference numerals are used for like elements in describing each drawing.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. Do not.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

1 shows a window used to determine the edge direction. Referring to FIG. 1, in a 3 × 3 window, the adjacent upper line is centered on x (i, j) , which is the interpolated pixel. Pixel value U _-1 , U ₀ , Pixels represented by U ₁ are located, and pixels represented by pixel values L- ₁ , L ₀ , and L ₁ are positioned in adjacent lower lines. Here, x (i, j) may represent the luminance value of the pixel, and the variable i, j may represent the row and column numbers, respectively. The Edge-based Line Average (ELA) method, which is widely used as an intra-field deinterlacing method, uses the edge-based correlation information on the direction between adjacent lines to determine the direction toward higher correlation. The pixels are interpolated with the average of two pixels around the corresponding direction. The correlation coefficient for the direction of the edge used in the ELA method may be expressed as in Equation 1.

Where C _-1 , C ₁ Is the diagonal direction, C ₀ is the vertical direction. That is, the direction of the edge is π / 4 in the direction C _-1, 3π / 4, C ₁ direction, the vertical direction of C, and ₀ can be expressed as, a high directivity in a direction to a minimum difference between the values of the two pixel correlation Seen to be determined in the direction of the edge. Then, the lost pixels are interpolated with the average value of the two pixels determined in the direction of the edge.

2 is a schematic diagram of the direction of the edge according to the M-ELA method.

Referring to FIGS. 1 and 2, the modified edge-based line average (M-ELA) method, which is an improvement of the ELA method in order to further improve image quality, is a method of interpolating pixels by finding a more accurate edge direction. Correlation information on the directionality of the edges by the M-ELA method is shown in the directions of P, Q, and V. FIG. The correlation coefficient for the P, Q, and V directions may be expressed by Equation 2.

M-ELA determines the directionality of the edge using the directions of P, Q and V represented by Equation 2 and the correlation coefficients of Equation 1. In the diagonal direction, the pixels are interpolated using an average of four adjacent pixels in the diagonal direction. If the correlation is highest in the P direction among the P, Q, and V directions (P value is the smallest), and C _-1 < C ₀ , the pixel value to be interpolated is ( U _-1 + U ₀ + L ₀ + L ₁ ) / 4. Similarly, the highest correlation (Q value is the smallest) in the Q direction among the P, Q, and V directions, and C ₁ If < C ₀ , the pixel value to be interpolated is ( U ₀ + U ₁ + L _-1 + L ₀ ) / 4. Lastly, in the remaining cases, the pixel value to be interpolated may be represented by ( U ₀ + L ₀ ) / 2.

Such a method of interpolating pixels using the M-ELA method can be expressed by Equation 3 below.

이다. 즉, M-ELA방식은 수직 및 대각선 방향의 두 화소간의 평균값에 1/2라는 가중치를 부여하여 합한 값으로 보간 예정화소인 x(i,j)를 나타낼 수 있고, 이는 두개의 화소의 평균값을 이용하는 2-탭 평균필터를 이용한 것이다. 그리고 수학식 3은 수학식 4와 같이 정리할 수 있다. here,

to be. That is, in the M-ELA method, the average value between two pixels in the vertical and diagonal directions is weighted 1/2 to represent the sum of the interpolated pixels x (i, j), which represents the average value of two pixels. The two-tap average filter used is used. Equation 3 may be arranged as in Equation 4.

The frequency response of the 2-tap average filter shows a semicircular shape, whereas the frequency response of the Sinc function is close to a rectangle. If the shape of the frequency response is close to a rectangle, the pixel can be effectively interpolated up to a high frequency region. Therefore, by using a filter having a frequency response shape close to a rectangle, the pixels can be interpolated more precisely. That is, in performing the deinterlacing, it is possible to interpolate pixels more precisely by using the filter using the Sinc function rather than the average filter. In other words, the pixel is interpolated by substituting the average value of the pixel using the 2-tap average filter with the value using a filter approximating the 6-tap Sinc function. In particular, the FDIF (Fixed Directional Interpolation Filter), which uses a fixed filter coefficient ( h (k) ), is used.

Accordingly, the present invention generates a candidate pixel value generator and a candidate pixel value generator for generating respective candidate pixel values corresponding to the directionality of the edges in the V direction perpendicular to the different diagonal directions P and Q directions. De-interlacing the pixels by applying weights based on the Wiener filter including a pixel interpolation unit applying a weight α calculated using a Wiener filter to each candidate pixel value and weighting them to interpolate the pixels. Provide the device. That is, the pixel X (i, j) to be interpolated may be expressed as a weighted value by applying an adaptive weight α to the candidate pixel value, and the weight α may be calculated using a Wiener filter. Can be.

FIG. 3 is a window illustrating a pixel column centering on an interpolation pixel to explain a method of generating candidate pixel values according to an exemplary embodiment of the present invention. Referring to FIG. 3, there are six reference pixels, each of which is diagonally or vertically adjacent to x (i, j) , which is an interpolation predetermined pixel. In addition, the pixel column displayed in gray is an area to be interpolated. After applying the filter coefficient h (k) fixed in the FDIF to the six reference pixels as the candidate weights h, the pixels are interpolated by weighting them.

For example, if the directionality of the edge is determined in the P direction, the interpolation predetermined pixel x (i, j) is x (i, j) = P ₀ h (0) + P ₁ h (1) + P ₂ h (2) + It can be represented as P ₃ h (3) + P ₄ h (4) + P ₅ h (5) . Also, if the directionality of the edge is determined in the Q direction, the interpolation predetermined pixel x (i, j) is x (i, j) = Q ₀ h (0) + Q ₁ h (1) + Q ₂ h (2) + Q ₃ h (3) + Q ₄ h (4) + Q ₅ h (5) .

If the fixed filter coefficient in the FDIF is h (k) , h (k) means a candidate weight value h applied to the reference pixel. That is, the present invention can assign an adaptive candidate weight value h in consideration of the distance and the correlation between the interpolation pixel and the adjacent reference pixels.

Accordingly, in the present invention, the interpolation predetermined pixel x (i, j) may be expressed as in Equation (5).

Here, P (k) represents the value of the reference pixel placed in the P direction, and represents pixel values of P ₀ , P ₁ , P ₂ , P ₃ , P ₄ , P _5, and the like, and Q (k) represents the Q direction. Is the value of the reference pixel placed at V , where V (k) represents the value of the reference pixel placed in the V direction (see FIG. 3). Also, P = (│ U _-1 -L ₀ │ + │ U ₀ - L ₁ │) / 2, Q = (│ U ₀ -L _-1 │ + │ U ₁ - L ₀ │) / 2, V = (│ U _-1 -L _-1 | + │ U ₀ - L ₀ | + │ U ₁ - L ₁ |) / 3. That is, the filter coefficient h (k ) is applied as the candidate weight value h to n adjacent reference pixels in each direction, and the candidate pixel value is generated by weighting the n adjacent pixels to which the candidate weight value h is applied.

That is, when the interpolation predetermined pixel x (i, j) has the highest correlation in the P direction and C ₋₁ < C ₀ , the filter coefficient h (k ) is respectively applied to each of n adjacent pixels in the P direction. Multiply weighted, with highest correlation in the Q direction, C ₁ If < C ₀ , each of n adjacent pixels in the Q direction may be multiplied and displayed by multiplying the filter coefficient h (k) . Therefore, the interpolation predetermined pixel x (i, j) can be expressed as shown in Equation (5). In addition, the filter coefficient used in the FDIF may be defined in various ways , but h (k) = {3, -17, 78, 78, -17, 3} / 128 may be applied. According to the present invention , a candidate pixel value can be extracted from the interpolation predetermined pixel x (i, j) represented by Equation (5 ) .

The candidate pixel generation unit for generating a candidate pixel value has an edge direction detection unit for detecting the directionality of the edges appearing in the V direction perpendicular to the P and Q directions which are different diagonal directions, and an edge direction detected by the edge direction detection unit. A fixed filter coefficient h (k ) is applied to the adjacent n pixels (where n is an integer) based on the FDIF (Fixed Directional Interpolation Filter) as a candidate weight value h, and the candidate weight value h ) And a candidate pixel value extraction unit for extracting candidate pixel values in the P, Q, and V directions by weighting the adjacent n pixels. Also, preferably n may be 6.

, V방향으로의 후보화소값은

및 Q방향으로의 후보화소값은

로 나타낼 수 있다.
A candidate pixel value in three directions can be extracted and generated from the interpolation predetermined pixel x (i, j) . That is, in x (i, j), which is the interpolation pixel, the candidate pixel value in the P direction is

, The candidate pixel value in the V direction is

And the candidate pixel value in the Q direction

.

A weight value α calculated using a Wiener filter is applied to each candidate pixel value along the P, Q, and V directions, and the pixel value X (i, j) to be interpolated can be obtained by weighted sum.

로 얻을 수 있다. 여기서 가중치(α)를 얻기 위해 위너필터(Wiener filter)를 이용한다. 이는 주변화소들도 후보화소값에 후보가중치(h)를 적용하고 가중합하여 이루어진다는 특성을 이용한 것이다. 또한, 평균제곱오차(mean square error)가 최소가 되도록 하는 위너필터(Wiener filter)는 통계적 성질을 나타내는 공분산 행렬을 이용하여 화소를 보간하는데 활용할 수 있다.
That is, the pixel value to be interpolated

Can be obtained. Here, a Wiener filter is used to obtain a weight α. This uses the characteristic that neighboring pixels are also made by applying and weighting candidate weights (h) to candidate pixel values. In addition, a Wiener filter for minimizing mean square error may be utilized to interpolate pixels using a covariance matrix representing statistical properties.

FIG. 4 illustrates peripheral pixels divided by window sizes WS to explain a deinterlacing apparatus according to an embodiment of the present invention. Referring to FIG. 4, when the window size WS is 1, U _-1 and U ₀ centering on the pixel X (i, j ) to be interpolated _. , U ₁ There are six peripheral pixels, labeled L _-1 , L ₀ , and L ₁ . Further, when WS = 2, U ² _-2 , U ² _-1 , based on the pixel X (i, j ) to be interpolated, U ² ₀ , U ² ₁ , U ² ₂ , L ² _-2 , L ² _-1 , L ² ₀ , L ² ₁ , L ² ₂ 10 peripheral pixels are indicated. Similarly, if WS = 3, 14 peripheral pixels can be located. That is, the peripheral pixels mean pixels located in the top row and the bottom row in the window centered on the pixel X (i, j) to be interpolated.

)을 생성하고, 행렬식

으로부터 가중치(α)를 연산하는 것을 특징으로 할 수 있다.
Peripheral pixels, which may be defined as the window size WS, may also be weighted by applying a weight α to the after-pixel value. Based on these characteristics, candidate pixel values in the P, Q, and V directions are extracted for each of the neighboring pixels, respectively.

), Determinant

It can be characterized by calculating the weight α from.

According to an embodiment of the present invention, when the window size WS is 1, a determinant for obtaining the weight α may be expressed as Equation 6.

Equation (6) is summarized as in Equation (7).

은 주변화소들의 화소값에 대한 행렬,

은 후보화소값의 행렬,

는 가중치이다. 즉,

은 각각의 주변화소에 대한 P, Q, V방향으로의 후보화소값으로 이루어진다. 따라서, WS=1일 때,

은 6×3의 행렬로 나타낼 수 있고, WS=2 일 때,

은 10×3의 행렬로 나타낼 수 있다. here,

Is a matrix of pixel values of neighboring pixels,

Is a matrix of candidate pixel values,

Is the weight. In other words,

Is a candidate pixel value in the P, Q, and V directions for each peripheral pixel. Therefore, when WS = 1,

Can be represented by a matrix of 6 × 3, and when WS = 2,

Can be represented by a matrix of 10 × 3.

In general, the larger the window size WS, the more accurate weight value α can be obtained because the weight value α is calculated based on more candidate pixel values, but the amount of calculation increases. Therefore, it is necessary to determine the appropriate window size WS.

When the weight α is obtained from Equation 7, Equation 8 is obtained.

를 구할 수 있다.
Thus, the pixel value to be interpolated

Can be obtained.

In calculating the weight α by applying the Wiener filter, various window sizes WS may be selected. In general, as the window size WS is increased, a more precise weight α can be calculated, but there is a problem that the amount of calculation increases. In addition, when the window size WS is a certain degree or more, the degree of improving the accuracy is reduced even when the window size WS is made larger. Therefore, it is necessary to adaptively select the window size WS.

5 is a flowchart illustrating a deinterlacing method according to an embodiment of the present invention. 5, in determining the window size (WS), utilizes the region characteristic coefficient (D _i). Area attribute coefficient (D _i) is a value representing the correlation between the pixels, to enlarge the window size (WS) until the value of D _i be larger than a preset value determines a window size (WS). If the value of D _i is small, it means that correlation is high among pixels, and if D _i is large, it means that correlation is low between pixels. That is, if the value of D _i is small and the correlation between pixels is high, the image quality will not deteriorate even if the window size WS is increased. If the value of D _i is large, the window size is low between the pixels. If (WS) is made small, the deterioration of image quality can be effectively prevented.

The present invention is characterized by calculating the area attribute coefficient (D _i) to the area attribute coefficient (D _i) to enlarge the window size until larger than a preset value, further comprising a window size determined for determining a window size A deinterlacing apparatus for interpolating pixels by applying a weight based on a Wiener filter is provided.

The region characteristic coefficient D _i is the value of four pixels ( U _i) in the leftmost and rightmost windows in the adjacent upper and adjacent lower lines of the pixel X (i, j) to be interpolated for each window size. , U _-i , L _i , L _-i ) and the values of the four pixels outside the nearest window in the row direction with respect to the four pixels in the window ( U _{i +1} , U- _{(i + 1)} , L _{i +1} , L- _{(i +1)} can be obtained by adding each difference value with). Referring to FIG. 4, when WS = 1, four pixels in the window are U ₋₁ , U ₁ , L _-1 , L ₁ , and the four pixels outside the window are U- ₂ , U ₂ , L _-2 , and L ₂ .

That is, the area attribute coefficient (D _i) can be calculated as shown in Equation (9). In addition, it is preferable to set four window sizes WS in Equation 9 by 1 ≦ i ≦ 3. This is because the calculation amount increases as the window size WS increases.

)을 생성하고, 행렬식

으로부터 가중치(α)를 연산할 수 있다. Referring to FIG. 4 and FIG. 5, the case starts when the window size WS is one. Starting with the case where the window size WS is 1 (i = 1), D _{1 is} calculated. If D ₁ is less than the preset value, i = i + 1 and D ₂ is calculated. If the calculated D ₂ is larger than the preset value, the window size WS is set to 2. If the calculated D ₂ is smaller than the preset value, the window size is increased to 3. This process is repeated until i <4. That is, the area attribute coefficient (D _i) to the window size (WS) for determining, and a candidate pixel value of the determined window size (WS) selected for the peripheral pixels based on the P, Q, V the direction of each of the peripheral pixels using To extract the candidate pixel values

), Determinant

The weight α can be calculated from.

Accordingly, the present invention generates a candidate pixel value generating step and a candidate pixel value generating step corresponding to each of the candidate pixel values corresponding to the directionality of the edges in the V direction perpendicular to the different diagonal directions P and Q directions. A deinterlacing method for applying interpolation to pixels by applying a Wiener filter-based weighting step comprising applying a weight α calculated by using a Wiener filter to a candidate pixel value and weighting and interpolating the pixels. to provide.

In addition, the de-interlacing method provided by the present invention enlarge the window size until it calculates the area attribute coefficient (D _i) larger than the area attribute coefficient (D _i) is a preset value, determines the window size to determine the window size The method may further include a step, wherein the region characteristic coefficient Di is a value of four pixels in the leftmost and rightmost windows in the adjacent upper and adjacent lower lines of the pixels to be interpolated for each window size and four in the window. The difference between the four pixel values outside the nearest window in the row direction on the basis of the pixels can be obtained by adding them.

In addition, the post-pixel value generating step may include detecting the directionality of the edges in the V direction perpendicular to the P and Q directions , which are different diagonal directions, and n adjacent n pixels in a direction based on the detected directionality of the edges. A fixed filter coefficient h (k ) is applied to the pixels as a candidate weight value h in a fixed directional interpolation filter (FDIF), and the weights of the adjacent n pixels to which the candidate weight value h is applied are added to the P, And extracting respective candidate pixel values for the Q and V directions.

FIG. 6 is a table illustrating objective performance of various deinterlacing schemes in terms of peak signal to noise ratio (PSNR). Referring to FIG. 6, it can be seen that the present invention has superior objective image quality compared to existing technologies. In particular, it can be seen that the average PSNR (peak signal to noise rate) in the case of applying the present invention is 0.89 dB and 0.91 dB higher than the FDD technique and the CAD technique, respectively. In addition, when the present invention is applied, it can be seen that the CPU time is reduced to 91.89% and 22.94%, respectively, compared to the FDD technique and the CAD technique.

FIG. 7 is a table illustrating evaluation of subjective picture quality of various deinterlacing schemes using MSSIM. SSIM (structure similarity) was used for the subjective image quality evaluation. SSIM is based on the structural similarity of two regions by region, and mean SSIM (MSSIM) is used as an objective index for the whole image. The MSSIM value ranges from 0 to 1, and the closer to 1, the more similar it is to the original. Referring to FIG. 7, it can be seen that the highest MSSIM value is obtained when interpolating pixels according to the present invention.

8 is a photograph showing a result of deinterlacing an image according to an embodiment of the present invention and a result of deinterlacing a conventional method. Referring to FIG. 8, the still image is partially enlarged to confirm how accurately the edge portion is interpolated. Where (a) Original Airplane, (b) Original partial image, (c) LA, (d) ELA, (e) EELA, (f) DOI, (g) NEDD, (h) LCID, (i) MELA, (j) FDD, (k) EMD, (l) ECA, (m) CAD, and (n) the proposed method.

As shown in the respective photographs, the deinterlacing apparatus and method for applying the Wiener filter-based weight to the candidate pixels of the present invention show an improved image compared to the conventional method.

The present invention can be embodied as computer readable programs or codes on a computer readable recording medium. Computer-readable recording media include all types of recording devices that store data that can be read by a computer system. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable program or code is stored and executed in a distributed fashion.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the present invention as defined by the following claims It can be understood that

Claims (12)

Area attribute coefficient (D _i) to the operation area to the characteristic factors (D _i) to enlarge the window size until larger than a preset value, the window size determiner for determining a first window size;
Based on the window size determined by the window size determiner, each candidate pixel value corresponding to directions of edges appearing in different diagonal directions P and Q directions and a vertical direction V direction are generated. A candidate pixel value generator; And
The candidate pixel value generated by the candidate pixel value generation unit applies a weight α calculated using a Wiener filter, and weights the candidate pixel value to which the weight α is applied to the pixels. A deinterlacing apparatus for interpolating pixels by applying a Wiener filter-based weight including a pixel interpolation unit to interpolate. delete The method of claim 1,
The candidate pixel value generation unit,
An edge direction detector for detecting the directionality of edges in the V direction perpendicular to the P and Q directions which are different diagonal directions; And
The filter coefficient h (k) fixed in the fixed directional interpolation filter (FDIF) is applied to adjacent n pixels (where n is an integer) in a direction based on the direction of the edge detected by the edge direction detector. h) and a candidate pixel value extraction unit configured to weight each of the adjacent n pixels to which the candidate weight value h is applied to extract candidate pixel values in the P, Q, and V directions. A deinterlacing device that interpolates pixels by applying filter-based weights. The method of claim 1,
The pixel interpolation unit,
A matrix of post-pixel values is extracted by extracting candidate pixel values in the P, Q, and V directions for each of the peripheral pixels located in the top row and the bottom row in the window centered on the pixel to be interpolated.

), Determinant

(here,

A deinterlacing apparatus for interpolating pixels by applying a Wiener filter-based weight, wherein the weight α is calculated from a matrix of pixel values of neighboring pixels. The method of claim 1,
The region characteristic coefficients D _i are four pixel values U _i , U _−i , L _i , L ₋ located at the leftmost and rightmost sides of the window in the upper line adjacent to the pixel to be interpolated for each window size and the lower line adjacent thereto. _i ) and four pixel values ( U _{i + 1} , U- _{(i + 1)} , L _{i + 1} , L- _{(i + 1)} located outside the nearest window in the row direction with respect to the four pixels in the window _. A deinterlacing apparatus for interpolating pixels by applying a Wiener filter-based weight, characterized in that the sum is obtained by adding the difference values with each other. The method of claim 5,
The area characteristic coefficient (D _i ) is D _i = │ U _i -U _{i + 1} │ + │ U _-i -U- _{(i + 1)} │ + │ L _i -L _{i + 1} │ + │ L _-i -L- _{(i + 1)} | De-interlacing apparatus for interpolating pixels by applying a Wiener filter-based weighting characterized in that. By computing the area attribute coefficient (D _i) to the area attribute coefficient (D _i) to enlarge the window size until larger than a preset value, the window size decision step of determining the window size;
Generating respective candidate pixel values corresponding to the directionality of the edges appearing in different diagonal directions P and Q and V directions perpendicular to the pixels to be interpolated based on the window size determined by the window size determination step ; Generating a candidate pixel value; And
A pixel calculated by using a Wiener filter is applied to each of the candidate pixel values generated in the candidate pixel value generating step, and the candidate pixel values to which the weight α is applied are weighted to add pixels. A deinterlacing method for interpolating pixels by applying a Wiener filter-based weighting step comprising interpolating a pixel. delete The method of claim 7, wherein
The candidate pixel value generation step,
Detecting the directionality of the edge appearing in the V direction perpendicular to the P and Q directions which are different from each other; And
A fixed filter coefficient h (k ) is applied as a candidate weight value h in a fixed directional interpolation filter (FDIF) to n adjacent pixels placed in a direction based on the detected direction of the edge, and the candidate weight value h Deinterlacing the pixels by applying a Wiener filter-based weighting method , comprising: weighting the adjacent n pixels to which the &lt; RTI ID = 0.0 &gt;) is applied &lt; / RTI &gt; . The method of claim 7, wherein
The weight α is a matrix of post-pixel values by extracting candidate pixel values in the P, Q, and V directions of peripheral pixels located in the top row and the bottom row in the window centered on the pixel to be interpolated. (

), Determinant

(here,

Is a matrix of pixel values of neighboring pixels). The method of claim 7, wherein
The region characteristic coefficients D _i are four pixel values U _i , U _−i , L _i , L ₋ located at the leftmost and rightmost sides of the window in the upper line adjacent to the pixel to be interpolated for each window size and the lower line adjacent thereto. _i ) and four pixel values ( U _{i + 1} , U- _{(i + 1)} , L _{i + 1} , L- _{(i + 1)} located outside the nearest window in the row direction with respect to the four pixels in the window _. A deinterlacing method for interpolating pixels by applying a Wiener filter-based weight, characterized in that each difference value is calculated by adding the difference value with The method of claim 11,
The area characteristic coefficient (D _i ) is D _i = │ U _i -U _{i + 1} │ + │ U _-i -U- _{(i + 1)} │ + │ L _i -L _{i + 1} │ + │ L _-i A deinterlacing method for interpolating pixels by applying a Wiener filter-based weight value, wherein -L- _{(i + 1)} |

Images