KR101402630B1 - Image interpolation method - Google Patents

Abstract

The present invention relates to an image interpolation method. The present invention provides an image interpolation method for computing luminance values of points that do not belong to a set of initial points to which luminance values are given, comprising: improving an image to remove artifacts resulting from compression of the image; Constructing a plurality of coordinate points for the interpolated image; Determining sixteen adjacent coordinate points on the low resolution images for each pixel position corresponding to the configured coordinate point and performing interpolation for each of the four directions based on the absolute value of the difference in brightness values in the four directions; Repeating the previous step for each color component of the image; And minimizing the artifacts and adjusting the edges of the image. Thereby, the artificial shadows generated due to the compression can be removed before the image is interpolated.

Description

Image Interpolation Method {IMAGE INTERPOLATION METHOD}

1 is a block diagram of a system for performing an image interpolation method according to the present invention,

2 is a flowchart illustrating an image interpolation method according to the present invention,

3 is a diagram illustrating an algorithm of an image interpolation method according to the present invention,

4 is a diagram illustrating an image interpolation process for an edge in the image interpolation method according to the present invention,

5 is a flowchart illustrating a background image processing process of the image interpolation method according to the present invention,

6 is a graph showing the brightness according to different values of the control variable n,

7 is a diagram illustrating an embodiment of an algorithm of the image interpolation method according to the present invention.

Description of the Related Art [0002]

101: processor 102: memory

103: Printer 104: Display

105: input device 106: data bus

The present invention relates to an image interpolation method. And more particularly, to an image interpolation method for removing artificial shading.

The image interpolation method is a basic operation in image processing. It is used routinely to rotate, warp, tightly combine, transform, and remove afterimages as well as enlarge and reduce images and increase resolution of video sequences . For example, interpolation is used when you want to resize images and video. It is also needed to generate full color images from CCD (Charge-Coupled Device) matrices that are widely used in digital cameras and camcorders and also affect general image processing. Numerous algorithms for image interpolation are proposed, and there is a large difference in complexity and performance between them.

If interpolation is performed by conventional methods such as a bilinear interpolation method and a bicubic interpolation method (Bicubic), artificial shading occurs in an area where there is a sudden change in brightness on the image. Typical artificial shadows appearing in the conventional interpolation method include checkerboard pattern in which an image is spread due to a jagged artificial shadow or a prediction of smoothing of the image, and a checkerboard pattern in which a circular artificial shadow (Ringing artifacts).

A typical two-dimensional interpolation method is performed by a series of one-dimensional interpolation along the mutually perpendicular direction. For example, the solution disclosed in U.S. Patent No. 6,915,026 proposes magnifying an image by two steps, namely, by first precomputing and storing the interpolation coefficients in advance in the column direction, then in the row direction . This technique is disadvantageous in that all of the one-dimensional interpolation directions are aligned in parallel with the image coordinate axis, and artificial shadows arise thereby. As a solution to this, it has been proposed to perform low-frequency filtering in advance in the diagonal direction, but this method degrades the quality of the image. U.S. Patent Nos. 5,719,967 and 6,606,093 disclose specific techniques for suppressing artifacts.

The solution disclosed in U.S. Patent No. 5,446,804 uses a pre-computed subpixel map for the edges. First, for each interpolated pixel, four adjacent pixels are specified. If there is no pixel to be cut from the rest by the outermost of the nearest neighboring pixels, the sum of the weights of the adjacent pixels is calculated. Then, for each pixel in which neighboring pixels are respectively present on the other side of the outer side, an intermediate value is calculated using two diagonal pixels existing on the same side of the outer side selected based on the edge map. Then, the value located on the other side is replaced with the value of the adjacent pixel located on the same side of the outer side and the calculated value. Such a solution produces an image with a sharp outline, but the image surface is displayed in a sealed form.

Furthermore, a solution is proposed to overcome the occurrence of stepping in the contour in performing interpolation according to the coordinate system arrangement and to perform image enlargement in two dimensions, i.e., vertical and horizontal directions at the same time.

The first solution is based on the assumption that deviations between low-resolution and high-resolution images are geometric binary. Storing the outline of an image is affected by the flexibility of the coefficient values needed for interpolation. This method performs magnification processing by a power of two. The interpolation is performed according to a two-step process. First, pixels existing in the coordinates of (2i + 1, 2j + 1) are calculated using known values existing at the positions of (2i, 2j). Then, the same method is performed by rotating the other pixels by 45 degrees. Each pixel is determined by a linear combination of four adjacent pixels, and the problem of determining the coefficients of the linear combination by the least square expression is solved. This interpolation method provides good results except in areas where the assumption about geometric duality is not applied.

The second solution is based on triangulation on the image, which considers the brightness value as the height of the point above zero level. The proposer of this solution constructs a mesh of triangles, linearly interpolates the intensity within the triangle, and reconstructs the image using the triangle's vertex values. This approach is based on Data-Dependent Triangulation (DDT), which combines optimization and standardization functions. According to this method, the triangulation method saves edges having bright brightness to enhance the image quality. However, this method is relatively complicated from the standpoint that the triangulation method is repeatedly performed.

Normally, enlargement processing of an image is performed in two steps. First, a luminance value is calculated, and background image processing is then performed to sharpen the image and improve the edge. For example, U.S. Patent No. 6,714,688 discloses a method for increasing the background image sharpness.

First, bilinear interpolation is used, and a method for processing the background image of the edge is applied in a window-based manner. The size of the window depends on the magnification of the image. First, low frequency filtering is applied to pixels inside the window to extract the high frequency components of the image. This high frequency component is added to the low frequency component together with a coefficient defined based on the local brightness level. The contour of the image is then further processed by the brightness conversion curve.

The solution disclosed in European patent EP 1533899A1 proposes a method of interpolating edges and further processing background images for resolution transformation of a digital image by the following steps. First, the image is enlarged to 2n times and then reduced to a desired magnification. The background image processing technique of the edge of the image is then applied. This solution provides good results, but there is a fairly complex problem. On the other hand, as a literature approach to the problem of image quality improvement, Li, X., and M. Orchard, "New edge direction interpolation method for image processing", IEEE International Conference on Vision at Vancouver, September 2000, IEEE Computer Graphics and Applications, Vol. 2, No. 3, 62-68P, entitled " Image reconstruction using data dependent triangulation ". All of these approaches have advantages, but they also have serious drawbacks.

Typically, the two-dimensional interpolation method successively performs one-dimensional interpolation in mutually perpendicular directions (see, for example, U.S. Patent No. 6,915,026). However, since the one-dimensional interpolation directions are all aligned in parallel with the image coordinate axis, their artificial shadows show the coordinate axis arrangement.

Here, there is a method that can prevent the occurrence of artificial shading by avoiding performing interpolation based on the axis arrangement, but these methods are somewhat complicated, and there is a problem that a lot of time and resources are required.

Improved methods for magnifying an image require a plurality of coordinate information of a fixed structure for magnifying an image by a power of 2 in particular. The problem shown in JPEG images or MPEG video sequences, etc. becomes more apparent by enlarging the image or by processing the background image in the background.

Accordingly, the present invention provides an image interpolation method capable of minimizing unexpected artifacts caused by eliminating the compression of an image file and the generation of cracks in the outline of a digital image while avoiding performing interpolation based on an axial arrangement .

The present invention also provides image interpolation that can represent the outline of an image with a higher resolution while minimizing the occurrence of artifacts by calculating image values at points that do not belong to the area of the initial points, And a method thereof.

The object is achieved by an image interpolation method comprising the steps of: improving an image to remove artifacts resulting from compression of an image; Constructing a plurality of coordinate points for the interpolated image; Determining sixteen adjacent coordinate points on the low resolution images for each pixel position corresponding to the configured coordinate point and performing interpolation for each of the four directions based on the absolute value of the difference in brightness values in the four directions; Repeating the previous step for each color component of the image; And minimizing the artifacts and adjusting the edges of the image.

Here, it is preferable to apply normal bipyte interpolation or edge direction interpolation for each pixel in accordance with edge information.

Then, it is preferable that a diagonal line is selected for the edge direction interpolation, the one-dimensional interpolation is performed first in the vertical and horizontal directions, and then the interpolation is performed at an arbitrary angle with respect to the image coordinate axes.

If the magnification R is used to enlarge the image one step at a time, the k-th initial image and the n-th intermediate images enlarged by T _k times in the case of T _k <R can be constructed.

In order to improve the image, the background image is processed to improve the local contrast by the brightness conversion curve, edge information E is extracted as the average differential value in the second direction, and the edge information E is extracted as the nonlinear It is desirable to apply the function.

Also, the image interpolation method preferably uses the following verification power function for 0 <d <1 and 0 <E <1.

Hereinafter, an image interpolation method according to the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram showing a configuration of an image processing apparatus for realizing an image interpolation method according to the present invention. 1, an image processing apparatus according to the present invention is operated by a processor 101 that executes a program code stored in a memory 102. As shown in Fig. In the memory 102, an initial value of a monochrome gray scale or a color image is stored. The image is analyzed and a new image having pixels calculated by the pixels of the original image is generated and transmitted to the display unit 104 or the printing apparatus 103. The flow of information within the image processing device is via the data bus 106. [

2 is a diagram illustrating an image interpolation method according to the present invention. Here, an original image having a size of 4 x 4 pixels or more is referred to as I. First, a set of coordinates necessary for calculating the brightness is calculated (S200). Then, for each point A, the following steps are performed. Here, the coordinates of the next coordinate A are calculated (S203), and 16 adjacent pixels for A belonging to the initial coordinate are determined (S204). Next, the absolute value of the differential result for the four directions is calculated by the following equation (S205). Here, the four pixels adjacent to A are referred to as I (i, j), I (i + 1, j), I (i, j + 1) and I (i + 1, j + 1).

[Equation 1]

Here, I _x , I _y , I _d1, and I _d2 are the absolute values of the differential results for the directions of the four directions, that is, the vertical, horizontal, and diagonal directions. Then, it is determined whether there is a detail image extending in the vertical or horizontal direction (S206). This is done by comparing I _x , I _y , I _d1 and I _d2 to each other. If I _x is less than I _d1 and I _x is less than I _d2 , or if I _y is less than I _d1 and I _y is less than I _d2 , that is, if there is a detail image in step S206, a bivariate interpolation is performed (S209).

If there is no detail image in step S206, one of two diagonal lines is selected (S207). More specifically, a diagonal line having a smaller differential value in the direction of I _d1 or I _d2 is selected, Lt; RTI ID = 0.0 &gt; (S208). &Lt; / RTI &gt; Step S208 will be described in more detail as follows. After the last point in the set of coordinates of the new image is processed, that is, step S202 must be satisfied. Steps S202 to S209 described above are repeated for all the color components in the RGB or YCrCb color space (S201). When step S201 is performed, the resulting image is output to the output device.

The method disclosed in the present invention can use interpolation in three steps. That is, the first two are arranged in image coordinates, and the third one forms an angle of 45 degrees with the image coordinate axis. The sharpness of the background image constitutes the local contrast enhancement in the edge region using the brightness transformation curve, and the degree of sharpness is controlled by the slope of the curve. As a result, the images are sharper and less jagged artificial shadows compared to conventional bipyramidal interpolation methods. When the kth image (k = 1, ..., n) is the initial image, the jagged artificial shadows constitute the nth intermediate images enlarged by Tk times when the enlargement of the image by the magnification R is performed by one step. By applying the proposed method in multiple ways. Where _Tk < R, and typically n = 2 is sufficient. At the same time, this method is very fast and does not require large memory capacity.

And, in the case of image enlargement, it is proposed to improve the image before interpolating to remove artificial shadows, which are burdened by lossy compression, such as block artifacts appearing in the case of the JPEG algorithm.

3 is a diagram showing the algorithm mentioned in step S208. The brightness of point A of a set of coordinates (indicated by crosses) surrounded by 16 adjacent pixels is required. It is assumed that a diagonal line connecting I (i, j) and I (i + 1, j + 1) is selected in step S207. At this time, a point close to the point A is selected from the point I (i + 1, j) and I (i, j + 1). (I (i + 1, j) in Fig. 3)

The points I (i, j), I (i + 1, j + 1) and I (i + 1, j) form a triangle. Points P ₁ and P ₂ are determined as above. Here, the straight line P ₁ P ₂ passes through the point A and is parallel to the selected diagonal. On the other hand, in order to more accurately determine the edge direction of the image, the inclination angle of the straight line P ₁ P ₂ can be changed by 45 degrees or 135 degrees. The brightness of points P ₁ and P ₂ is determined by one-dimensional cubic interpolation and is obtained by using four values in the vertical direction with respect to P ₁ and four values in the horizontal direction with respect to P ₂ .

To obtain the brightness of point A, a linear interpolation between points P ₁ and P ₂ is used. Here, the one-dimensional interpolation method is not limited to the three-dimensional or linear interpolation, and it is possible to use a different interpolation method and a large number of grid points to calculate the brightness of the point A of the other points P ₁ and P ₂ .

4 is a view illustrating a process of selecting a diagonal line. As shown in FIG. 4, if the edge truncates the vertex I ₃ , an inequality of | I ₂ -I ₄ | <| I ₁ -I ₃ | holds for the brightness of the vertex. And point I ₃ is not used to calculate the brightness of the point marked "?".

One-dimensional cubic interpolation can be performed quickly if the grid is regular and the inter-pixel distance of the original image I is equal to one. If the initial values at position [-1,0,1,2] in the cubic interpolation are υ ₁ , υ ₂ , υ ₃ , υ ₄ , the value υ _△ at the point where △ is greater than 0 and less than 1 should be obtained . υ _Δ is obtained by the following equation.

&Quot; (2) &quot;

If υ ₁ , υ ₂ , υ ₃ , υ ₄ are integers, C ₁ and C ₂ can be computed using the integral method. Next, background image processing is performed as needed to emphasize and sharpen the edges.

FIG. 5 shows the basic steps for increasing the resolution of the edge, i. E., Correcting the edge of the digital image. First, the current edge thicknesses Kv and Kh in the horizontal and vertical directions are determined (S500). Here, Kv and Kh may be values which have been truncated according to degree of image enlargement in the horizontal and vertical directions. At this time, in order to enlarge the local contrast, a look-up table of the brightness conversion curve is referred to (S501).

The brightness conversion curve exists in the segment [0,1] and the derivative value at point 0.5 should be 1 or more. The resulting sharpness of the edge depends on the shape of the curve, in particular on the derivative value at 0.5. That is, the larger the derivative value, the more the sharpness of the edge increases. The function of the brightness conversion curve can be expressed by the following equation.

&Quot; (3) &quot;

Here, the derivative of the curve at x = 0.5 is adjusted by the parameter n.

6 is a diagram showing various types of brightness conversion curves according to the parameter n. Each pixel of the original image is processed according to the following procedure. First, the channel c1, which is the brightest among the three channels of the color space RGB or YCrCb, is determined. Further, the brightness of _Ic1 (i, j) corresponding to the channel given in steps S504 to S508 is processed. If the minimum and maximum values of the brightness are obtained in the adjacent pixels, the input pixel I _c1 (i, j) is scaled in the segment [0,1] as shown in the following equation (S505).

&Quot; (4) &quot;

Then, d is obtained from the brightness conversion curve in step S506, and scaled by the segment [local_minimum, local_maximum] is as follows.

&Quot; (5) &quot;

In step S507, E indicating the sharpness of the edge is obtained. Assuming that K2h is the value at which Kh / 2 is discarded at the end, and K2ν is the value at Kν / 2, the value of E is deducted from the following equation.

&Quot; (6) &quot;

In step S508, d is converted by the following equation under the assumption that the pixel values change from 0 to 255:

&Quot; (7) &quot;

는 배경 이미지 처리 후 컬러 성분의 최종값이다. 이미지가 하나 이상의 컬러 성분을 가지는 경우에, 나머지 성분들은 컬러 성분과

및 컬러 채널 I_c1(i,j)의 참조값의 비의 곱으로 구한다.here,

Is the final value of the color component after background image processing. If the image has one or more color components,

And the reference value of the color channel I _c1 (i, j).

&Quot; (8) &quot;

As described above, the color channels c1 is selected by the maximum value of _{I c1 (i, j),} I c2 (i, j), I c3 (i, j). If I _c1 (i, j) is 0, all the remaining color components are also zero.

FIG. 7 is a diagram for enlarging one area of an image five times according to the bipartite interpolation method and the interpolation method according to the present invention.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments.

As described above, according to the image interpolation method according to the present invention, unexpected artificial shadows generated due to the compression of the image file and the generation of cracks in the outline of the digital image while avoiding interpolation based on the axis arrangement Can be minimized.

According to the image interpolation method of the present invention, by calculating the brightness values at the points not belonging to the area of the initial points having the brightness value, the outline of the image can be expressed with a higher resolution while minimizing the occurrence of artificial shading .

Claims (6)

In the image interpolation method, Calculating a coordinate point for calculating the brightness of the image; Determining a pixel corresponding to sixteen adjacent coordinates for a pixel corresponding to the calculated coordinate point; Calculating a difference of brightness values between pixels corresponding to the determined adjacent coordinates; Performing interpolation of a pixel corresponding to the calculated coordinate point and a pixel corresponding to the adjacent coordinate based on the absolute value of the calculated value; Repeating the previous steps for each color component of the image; And minimizing artificial shadows and correcting the edges of the image. The method according to claim 1, And performing interpolation or edge direction interpolation of a pixel corresponding to the calculated coordinate point based on information on an edge. 3. The method of claim 2, Wherein a diagonal line is selected for the edge direction interpolation, the one-dimensional interpolation is performed first in the vertical and horizontal directions, and then the interpolation is performed at an arbitrary angle with respect to the image coordinate axes. The method according to claim 1, Wherein the k-th initial image and the n-th intermediate images enlarged by T _k times in the case of T _k < R are constituted when enlargement of the image is performed by the magnification R by one step. The method according to claim 1, In order to correct the edge of the image, the background image is processed to improve the local contrast by the brightness conversion curve, and information about the edge is extracted as the average differential value in the horizontal and vertical directions to determine the current pixel value d and the sharpness And a non-linear function is applied according to E. 5. The method of claim 4, The image interpolation method includes: Wherein the following power function is used for the case where 0 <d <1 and 0 <E <1.

Images