CN103500437A - Fast NEDI (new edge-directed interpolation) image interpolation implementation method - Google Patents

Abstract

The invention discloses a fast NEDI image interpolation realization method. The steps of the method are: judge whether the magnification factor n is 0, if yes, jump to step five, if not, continue; use a template to calculate the difference coefficient, and the coefficient can be It is divided into the difference coefficient A1 of the center point and the difference coefficient B1 of the non-center point; repeatedly use the coefficients of A1 and B1 and the known pixel points to calculate the pixel value with interpolation points, and obtain a 2 times enlarged image; judge the zoom Whether the multiple n is 1, if yes, go to step five, if not, n=n-1 and go to step three; end, output the final image. The invention adopts a circular template, which can better reflect the local statistical information of the low-resolution image, improve the accuracy of the interpolation coefficient, eliminate the ringing phenomenon near the edge after interpolation, and obtain a high-resolution image with clearer and sharper edges; Moreover, the interpolation coefficients are only calculated once and reused in the high-magnification image enlargement process, which greatly reduces the computational complexity.

Description

A Fast Implementation Method of NEDI Image Interpolation

technical field

The invention belongs to the field of NEDI algorithms, in particular to a fast NEDI image interpolation realization method.

Background technique

Super-resolution technology has its application value in many fields, including high-definition television, high-resolution printing of video images, medical imaging, aviation and satellite imaging, remote sensing, surveillance images, and digital photography. However, in the process of image acquisition, due to the limitations of various factors such as imaging system (such as CCD camera, etc.), external environment, and imaging technology, the image has different degrees of degradation, which is difficult to meet the actual needs. Enlarging digital images is an important topic in multimedia technology and image display technology. Existing image scaling algorithms are roughly divided into traditional interpolation algorithms and edge adaptive algorithms. The former includes nearest neighbor interpolation, bilinear interpolation, and cubic convolution interpolation. Because they use the same interpolation function for the entire image, the scaled image will have blockiness or blurred details.

Therefore, many scholars at home and abroad are committed to finding new interpolation methods that can obtain better visual effects. These methods first need to obtain accurate edge direction information of the image and then use different interpolation methods. However, accurately estimating the edge direction is still a problem to be solved, so these algorithms will inevitably cause more artificial processing traces in the interpolated image, which will affect the visual quality. In order to solve this problem, Li and Orchard proposed a NEDI (NewEdge-Directed Interpolation)[5-6] algorithm in 2001 to interpolate unknown pixels in high-resolution images by using the intrinsic local structural properties and statistical properties of low-resolution images. This method is a typical algorithm with good edge-preserving properties. Compared with traditional linear interpolation, it avoids the degradation of edge details caused by cross-edge interpolation, and significantly improves the visual quality of the image.

其中y=[y₁...y_k...y_M ²]^T为一个包含M×M个点像素的数值向量，这M×M个像素包含在一个局部窗中；C为4×M²数值矩阵，其第k列向量为y_k对角线方向的最近邻4点像素值。The classic NEDI algorithm uses an M×M rectangular template centered at the point to be interpolated when estimating the local covariance, and uses the formula when calculating the covariance coefficient at low resolution

Among them, y=[y ₁ ...y _k ...y _M ² ] ^T is a numerical vector containing M×M pixels, and these M×M pixels are included in a local window; C is 4×M ^2Numerical matrix, whose kth column vector is the pixel value of the nearest 4 points in the diagonal direction of y _k .

Although the calculation of the rectangular template is simple, it does not have directionality and cannot take into account the direction of the pixel edge in the template, so the geometric characteristics of the local area cannot be accurately estimated, and this is exactly what causes the edge part of the image to vibrate after interpolation by the classic NEDI algorithm. The main cause of the bell effect. And the classic NEDI adopts the M×M rectangular template based on the assumption that all pixels in the template have the same contribution to the calculation of α. However, in fact, because the distance from each pixel in the template to the center point to be interpolated is inconsistent, the degree of influence is not the same. , so this assumption is often not valid.

Using the classic NEDI algorithm to enlarge the image twice requires two steps. The first step is to interpolate the center point of the original four pixels, and the second step is to interpolate the pixels in the horizontal and vertical directions based on the interpolated center point pixel. . In the classic NEDI algorithm, the second-step interpolation will use the pixels generated in the first step, which may cause error transmission and affect the final interpolation result. The algorithm proposed in the past literature uses 6 adjacent downsampling points around the point to be interpolated when performing the second step of interpolation, which can eliminate the problem of error accumulation, but the calculation of interpolation coefficients is inaccurate and the high and low resolution covariance cannot always satisfy the geometric requirements. The flaw of duality remains unresolved.

Using the classic NEDI algorithm to enlarge the image twice requires two steps. The first step is to interpolate the center point of the original four pixels, and the second step is to interpolate the pixels in the horizontal and vertical directions based on the interpolated center point pixel. . When the image is enlarged by four times, the same process is repeated, and since the number of pixels used this time is four times that required before, the amount of calculation will increase exponentially.

The classic NEDI algorithm uses a rectangular template when calculating interpolation coefficients, which cannot take into account the directionality of pixels in the template, and only uses a single window during interpolation, which may cause the covariance of high and low resolutions to fail to satisfy geometric duality, resulting in obvious edges of the image after interpolation The ringing phenomenon affects the visual quality. Moreover, the classic NEDI algorithm has high computational complexity, which limits the practical application of this method.

Although NEDI realizes direction adaptation, its computational complexity is very high. Since edge pixels often only occupy a small part of an image, a new algorithm is needed to reduce the computational complexity.

Contents of the invention

The purpose of the embodiments of the present invention is to provide a fast NEDI image interpolation implementation method, aiming at solving the problem that the classic NEDI algorithm has high computational complexity, which limits the practical application of the method.

The embodiment of the present invention is realized like this, a kind of fast NEDI image interpolation realization method, this method comprises the following steps:

Step 1: Determine whether the magnification factor n is 0, if yes, go to step 5; if not, continue;

Step 2: Use the template to calculate the difference coefficient, which can be divided into the difference coefficient A1 of the central point and the difference coefficient B1 of the non-central point;

Step 3: Repeatedly use the coefficients of A1 and B1 and the known pixel points to calculate the pixel value with interpolation points, and obtain a 2 times enlarged image;

Step 4: Determine whether the magnification factor n is 1, if yes, go to step 5; if not, n=n-1 and go to step 3;

Step 5: End, output the final image.

Further, the improved interpolation template is a circular template whose center is located at the point to be interpolated.

Further, the strategy of reusing the interpolation coefficients takes 4 times magnification of an image as an example.

Further, the image is magnified by 4 times, that is, several points are to be inserted in the following four original pixel points,

To achieve 4 times image magnification, that is, two times of magnification are required. In the first 2 times of magnification: a set of interpolation coefficients A1 is obtained by center point interpolation; a set of interpolation coefficients B1 is obtained by non-center point interpolation;

According to the regional similarity, during the second 2x zoom-in: as long as the center point to be interpolated is within the area of these 4 original pixels, the group of coefficients A1 can be used repeatedly for interpolation calculation. At this time, the center point interpolation will be used One original point, one first interpolated center point, two first interpolated non-central points, and interpolation coefficient A1;

The second non-central point interpolation uses the same idea to repeatedly use the B1 set of coefficients.

The present invention adopts the mixed interpolation method, first utilizes the sobel operator to detect the edge, divides the original pixel into two areas: flat area and edge area, carries out bicubic interpolation for the flat area pixel; adopts the improved method of the present invention for the edge area NEDI algorithm; the fast NEDI image interpolation implementation method provided by the present invention adopts a circular template, which can better reflect the local statistical information of the low-resolution image, improve the accuracy of the interpolation coefficient, and can eliminate most of the edges near the edge after interpolation. Ringing phenomenon, resulting in high-resolution images with clearer and sharper edges. Moreover, the interpolation coefficients are only calculated once and reused in the high-magnification image enlargement process, which greatly reduces the computational complexity.

Description of drawings

Fig. 1 is the flow chart of the fast NEDI image interpolation realization method that the embodiment of the present invention provides;

Fig. 2 is a schematic diagram of image interpolation algorithm processing provided by an embodiment of the present invention;

Fig. 3 is a schematic diagram of a circular template provided by an embodiment of the present invention;

Fig. 4 is a schematic diagram of a 4-fold image magnification provided by an embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention more clear, the present invention will be further described in detail below in conjunction with the examples. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

Fig. 1 shows the flow of the fast NEDI image interpolation implementation method provided by the present invention. For ease of illustration, only the parts relevant to the present invention are shown.

The fast NEDI image interpolation realization method that the embodiment of the present invention provides comprises the following steps:

Step 1: Determine whether the magnification factor n is 0, if yes, go to step 5; if not, continue;

Step 2: Use the template to calculate the difference coefficient, which can be divided into the difference coefficient A1 of the central point and the difference coefficient B1 of the non-central point;

Step 3: Repeatedly use the coefficients of A1 and B1 and the known pixel points to calculate the pixel value with interpolation points, and obtain a 2 times enlarged image;

Step 4: Determine whether the magnification factor n is 1, if yes, go to step 5; if not, n=n-1 and go to step 3;

Step 5: End, output the final image.

As an optimization solution of the embodiment of the present invention, the improved interpolation template is a circular template whose center is located at the point to be interpolated.

As an optimization solution of the embodiment of the present invention, the interpolation coefficient reuse strategy takes 4 times zooming in on an image as an example.

As an optimization scheme of the embodiment of the present invention, the image is magnified by 4 times, that is, several points must be inserted in the following four original pixel points,

To achieve 4 times image magnification, that is, two times of magnification are required. In the first 2 times of magnification: a set of interpolation coefficients A1 is obtained by center point interpolation; a set of interpolation coefficients B1 is obtained by non-center point interpolation;

According to the regional similarity, during the second 2x zoom-in: as long as the center point to be interpolated is within the area of these 4 original pixels, the group of coefficients A1 can be used repeatedly for interpolation calculation. At this time, the center point interpolation will be used One original point, one first interpolated center point, two first interpolated non-central points, and interpolation coefficient A1;

The second non-central point interpolation uses the same idea to repeatedly use the B1 set of coefficients.

The application principle of the present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

The present invention adopts the hybrid interpolation method, as shown in Fig. 2, first uses the sobel operator to detect the edge, and divides the original pixel into two regions: a flat region and an edge region. Bicubic interpolation is carried out for the pixels in the flat area; for the edge area, the improved NEDI algorithm in this paper is used.

As shown in Figure 1, the improved NEDI algorithm provided by the present invention carries out the specific steps of the flow process of the algorithm of 2n times amplification as follows:

S101: Determine whether the magnification factor n is 0? If yes, go to Step5; if not, continue

S102: Use the template proposed in this paper (will be elaborated in Section 2.2.1) to calculate the difference coefficient, which can be divided into the difference coefficient A1 of the central point and the difference coefficient B1 of the non-central point

S103: Repeatedly use the coefficients of A1 and B1 and the known pixel points to calculate the pixel value with interpolation points (the specific method will be explained in Section 2.2.2), and obtain a 2 times enlarged image

S104: Determine whether the magnification factor n is 1? If yes, go to Step5; if not, n=n-1 and go to Step3

S105: end, and output the final image.

Through improvement, the method can obtain accurate center point interpolation coefficients and non-center point interpolation coefficients by one parallel calculation. By using these coefficients and the original pixels, a 2x (or even higher) enlarged image can be obtained.

The improved interpolation template provided by the present invention is as follows:

In the NEDI algorithm, the interpolation coefficient will determine the interpolation result of the center point pixel in the first step, but for the improved algorithm, since this set of interpolation coefficients will be reused in the subsequent high-magnification zoom-in, the accuracy of the interpolation coefficients is very high. important. The classic NEDI algorithm uses a rectangular template whose center is located at the point to be interpolated when estimating the local covariance. Although the calculation of this template is simple, it does not have directionality and cannot take into account the direction of the edge of the pixel in the template, so it cannot be accurately estimated. The geometric characteristics of the local area, and this is precisely the main reason for the ringing effect at the edge of the image after interpolation by the classic NEDI algorithm. And the classic NEDI adopts the M×M rectangular template based on the assumption that all pixels in the template have the same contribution to the calculation of α. However, in fact, because the distance from each pixel in the template to the center point to be interpolated is inconsistent, the degree of influence is not the same. , so this assumption is often not valid.

In order to solve this problem, the present invention proposes a circular template whose center is located at the point to be interpolated to calculate the interpolation coefficient. The schematic diagram of the template is shown in FIG. 3 (the template radius is represented by MT). On the one hand, the use of the circular template eliminates the interference of the four corner points in the original rectangular template on the estimation of the interpolation coefficient, and the circular template can take into account the directionality of the pixels, thereby improving the accuracy of the calculation of the interpolation coefficient.

The interpolation coefficient reuse strategy provided by the present invention:

In order to reduce computational complexity, the present invention proposes a strategy for reusing interpolation coefficients when performing high-magnification amplification. Take the 4 times magnification of the image as an example, that is, to insert several points in the following four original pixel points, as shown in Figure 4,

To achieve 4 times image magnification, two times of 2 times magnification are required. During the first 2 times magnification: a set of interpolation coefficients A1 is obtained by center point interpolation; a set of interpolation coefficients B1 is obtained by non-center point interpolation. Then, according to the regional similarity, during the second 2x zoom-in: as long as the center point to be interpolated is within the area of these 4 original pixels, the group of coefficients A1 can be used repeatedly for interpolation calculation (at this time, the center point interpolation will be Use an original point, a central point inserted for the first time, two non-central points inserted for the first time, and the interpolation coefficient A1); the second non-central point interpolation is the same idea and reuse the B1 group coefficient. Since the NEDI algorithm’s largest calculation load is in the calculation of interpolation coefficients, and after obtaining the interpolation coefficients, the calculation load of the interpolation itself is not large, and the improved algorithm finally realizes 4 times the size of the image (or even Higher magnification), so the total calculation amount of the system is greatly saved.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention should be included in the protection of the present invention. within range.

Claims (4)

1. a quick NEDI image interpolation implementation method, is characterized in that, the method comprises the following steps:

Step 1: judge that whether enlargement factor n is 0, if so, jumps to step 5; If non-, continue;

Step 2: use formwork calculation difference coefficient, this coefficient can be divided into the difference coefficient B 1 of difference coefficient A1 and the non-central point of central point;

Step 3: reuse this cover coefficient of A1 and B1 and known pixels point and calculate the pixel value with interpolation point, obtain 2 times of enlarged images;

Step 4: judge that whether enlargement factor n is 1, if so, jumps to step 5; If non-, n=n-1 also jumps to step 3;

Step 5: finish the output final image.

2. quick NEDI image interpolation implementation method as claimed in claim 1, is characterized in that, this improved interpolation template is a kind of circular shuttering that is centered close to interpolation point.

3. quick NEDI image interpolation implementation method as claimed in claim 1, is characterized in that, this interpolation coefficient is reused strategy and is enlarged into example so that image is carried out to 4 times.

4. quick NEDI image interpolation implementation method as claimed in claim 3, is characterized in that, image carries out 4 times of amplifications, will insert out several points in four original image vegetarian refreshments below;

Realize that 4 times of images amplify, will carry out twice 2 times of amplifications, when 2 times of amplifications for the first time: the central point interpolation has obtained one group of interpolation coefficient A1; Non-central point interpolation has obtained one group of interpolation coefficient B1;

According to regional similarity, during 2 times of amplifications for the second time: as long as central point to be inserted is to carry out interpolation calculation with regard to this group coefficient of reusable A1 in the zone of these 4 original pixels, now, the central point interpolation will use an original point, a central point of inserting out for the first time, two non-central points of inserting out for the first time, and interpolation coefficient A1;

Non-central point interpolation is that same thinking is reused this group coefficient of B1 for the second time.

Images