CN106169174B - Image amplification method - Google Patents
Image amplification method Download PDFInfo
- Publication number
- CN106169174B CN106169174B CN201610509556.2A CN201610509556A CN106169174B CN 106169174 B CN106169174 B CN 106169174B CN 201610509556 A CN201610509556 A CN 201610509556A CN 106169174 B CN106169174 B CN 106169174B
- Authority
- CN
- China
- Prior art keywords
- image
- gradient
- interpolation
- resolution
- similarity
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 70
- 230000003321 amplification Effects 0.000 title abstract description 8
- 238000003199 nucleic acid amplification method Methods 0.000 title abstract description 8
- 230000000694 effects Effects 0.000 claims abstract description 10
- 238000012937 correction Methods 0.000 claims abstract description 8
- 238000001914 filtration Methods 0.000 claims abstract description 6
- 230000002093 peripheral effect Effects 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 238000000605 extraction Methods 0.000 claims description 6
- 239000000126 substance Substances 0.000 claims description 6
- 238000005259 measurement Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000012986 modification Methods 0.000 claims description 2
- 230000004048 modification Effects 0.000 claims description 2
- 102100037651 AP-2 complex subunit sigma Human genes 0.000 claims 2
- 101000806914 Homo sapiens AP-2 complex subunit sigma Proteins 0.000 claims 2
- 238000012805 post-processing Methods 0.000 abstract description 5
- 230000006870 function Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000006731 degradation reaction Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000015556 catabolic process Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000004321 preservation Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000011524 similarity measure Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Editing Of Facsimile Originals (AREA)
Abstract
The invention discloses an image amplification method, which comprises the following steps: a gradient value estimation step, wherein the gradient of the high-resolution image is estimated by adopting a non-local mean filtering method; a gradient guiding interpolation step, wherein the interpolation of the edge pixel is guided by using the gradient value; an interpolation result correction step, wherein a non-local mean filtering method used in a gradient estimation stage is adopted to carry out post-processing on an interpolation image, and noise and artificial effect introduced by interpolation are removed; and a texture structure reconstruction step, wherein the interpolated image is used as an initial result, the gradient of the interpolated image is used as a constraint, and a reconstruction-based method is adopted to obtain a final high-resolution image.
Description
Technical Field
The invention relates to the technical field of video image processing, in particular to an image amplification method based on structure maintenance.
Background
With the popularization of digital products, images are increasingly widely used as main sources for people to obtain information. At the same time, digital image processing techniques have also been rapidly developed. The acquisition of video images is a key step in digital image processing systems. In the digital acquisition process, the image resolution and image quality are degraded by several factors. The sampling frequency, undersampling, causes aliasing of the spectrum of the image, and degradation occurs due to the deformation effect. Atmospheric disturbances, defocus, sensor size, and relative motion between the image capture device and the object being photographed can cause blurring of the image. Noise, such as gaussian noise, is also introduced during the acquisition, transmission and storage of the image, and the image is degraded.
Therefore, how to improve the resolution and quality of an image to be as close to the original image as possible has become one of the research hotspots in the image processing field in the international recent years. With the development of image processing technology and the continuous improvement of computer computing capability, the super-resolution reconstruction technology of video images provides a good solution for reconstructing low-resolution images. The method can enlarge a series of low-resolution images according to a certain proportion to finally generate one or more high-resolution images, and well maintain the structure of the original image.
The existing super-resolution reconstruction methods are mainly divided into three categories. The first category is super-resolution techniques based on interpolation. The second category is reconstruction-based super-resolution techniques. The third category is learning-based super-resolution techniques. Simple linear interpolation techniques, such as bilinear and bicubic interpolation, are computationally simple but produce a jagged effect, while also blurring the edges. In order to better maintain the sharpness of the edge, many edge-guided interpolation methods are proposed one after another. Researchers in 2001 proposed to estimate the covariance of the high-resolution image on the low-resolution image and then use the covariance for interpolation. Another researcher proposed a block-based autoregressive model in 2008, which estimates an entire block of pixels at a time. Another researcher proposed a robust soft-decision interpolation technique in 2012, which employs a weighted least squares method in the estimation of both parameters and pixels. Then, these methods consider only the reconstruction of the edge portion, and do not consider the reconstruction of the texture portion.
The super-resolution technology based on reconstruction is an inverse process of simulating image degradation to solve an optimization equation. The image degradation process is to obtain a low-resolution image by down-sampling a high-resolution image after blurring. Another method based on total variation of images proposed by researchers in 2005 is a very representative one of such methods. In this method, the total variation of the image is added as a constraint term to the optimization equation, thereby constraining the solution to the problem. It can greatly suppress the artifacts while maintaining edge sharpness. Another researcher proposed in 2011 to estimate the gradient of the edge of the high-resolution image by the gradient of the low-resolution image, and then add the estimated gradient to the optimization equation as a constraint term.
In recent years, some super-resolution reconstruction methods based on learning are also continuously proposed. Another researcher proposed a super-resolution reconstruction method based on sparse representation in 2010. The method provides that the image blocks can be represented by elements in an overcomplete dictionary in a linear combination manner, wherein the number of non-zero coefficients is as small as possible. Therefore, two overcomplete dictionary sets are first generated, where the image blocks in the two sets are in one-to-one correspondence, i.e., low resolution images and high resolution images, respectively. For any low-resolution image block input, a sparse representation is found in the low-resolution dictionary, and then the high-resolution image block is generated in the high-resolution dictionary by using the set of sparse representations. Another investigator proposed in 2016 to use a deep learning approach to reconstruct high resolution images. The basic method is to generate a plurality of sets of low-resolution and high-resolution image pairs, then train the network by taking the low-resolution images as the input of the convolutional neural network and the high-resolution images as the output of the convolutional neural network. For a trained network, any low resolution image is taken as input, and a high resolution image is generated as a reconstruction result.
The prior art algorithm is based on interpolation, and although the calculation amount is low, the reconstruction effect is poor. The reconstruction-based method does not reconstruct both the edge and texture parts well at the same time. The learning-based method has high computational complexity and strong dependence on the selection of a training library.
The invention provides an image amplification method based on structure preservation, which can well reconstruct the edge structure and the texture structure of an image. Firstly, the gradient of the high-resolution image is estimated by adopting a modified non-local mean method, and then the interpolation of the edge pixel is guided by the gradient. And then, carrying out post-processing on the interpolation image by adopting a non-local mean method used in a gradient estimation stage, and removing noise and artificial effect introduced by interpolation. And finally, taking the interpolated image as an initial result, taking the gradient of the interpolated image as constraint, and obtaining a final high-resolution image by adopting a reconstruction-based method.
Disclosure of Invention
The invention discloses an image amplification method, which comprises the following steps: a gradient value estimation step, wherein the gradient of the high-resolution image is estimated by adopting an improved non-local mean method; a gradient guiding interpolation step, wherein the gradient value is used for guiding the interpolation of the edge pixel; an interpolation result correction step, wherein a non-local mean method used in a gradient estimation stage is adopted to carry out post-processing on an interpolation image, and noise and artificial effect introduced by interpolation are removed; and a texture structure reconstruction step, wherein the interpolated image is used as an initial result, the gradient of the interpolated image is used as a constraint, and a reconstruction-based method is adopted to obtain a final high-resolution image.
In the image enlargement method of the present invention, the gradient value estimation step includes a similarity measurement step and a high-resolution gradient estimation step.
In the image amplifying method of the present invention, in the similarity measuring step, the similarity of the pixel points is defined by the similarity of the image blocks, wherein, assuming that the current pixel point is y (i, j), the image block composed of the peripheral N × N pixel points is N (i, j), and assuming that another pixel point in the image is y (m, N), the image block composed of the peripheral N × N pixel points is N (m, N), the similarity between the pixel points y (i, j) and y (m, N) is estimated by the gray level intensity similarity of the corresponding image blocks.
In the image enlarging method of the present invention, the difference in gray-scale intensity between the image blocks is defined by formula (1):
wherein the content of the first and second substances,is an operator of the second normal form,
a weight is given to the pixel point y (m, n) to measure the similarity, as shown in formula (2):
where Z (i, j) is a normalized constant representing the sum of all weights, and the parameter σ 1 controls the decay rate of the exponential equation.
In the image amplification method, in the high-resolution gradient estimation step, a low-resolution image is processed by using the traditional bicubic interpolation to obtain an initial high-resolution image, and then the initial high-resolution image is subjected to convolution operation by using a Sobel operator to obtain the gradient approximate estimation of the high-resolution image.
In the image enlargement method of the present invention, the gradient correction is as shown in formula (3):
wherein the window sxs size is set to 21 × 21.
In the image enlarging method of the present invention, in the gradient guide interpolation step, the plane is divided into 8 directional sections.
Subdividing the 8 directions into two categories, omega respectively1＝{β1,β3,β5,β7And Ω2＝{β0,β2,β4,β6}. Suppose the gradient direction of Mi is θMIf theta is greater than thetaMBelong to omega1As shown in fig. 5, the black point is projected to the gradient direction, and assuming that p (ij) is the projection length, the weight of Nj is defined as formula 4:
where c (i) is a normalization parameter representing the sum of all weights, σ 2 controls the decay rate of the exponential equation.
In the image enlargement method of the present invention, the weight on the side with the larger gray scale difference is set to 0, where σ 2 is 0.2 and the threshold T is 50.
In the image enlargement method of the present invention, in the texture reconstruction, it is assumed that the input low-resolution image is IlThe interpolated image isThe super-resolution technology based on reconstruction is adopted, the gradient of the interpolation image is used as a constraint condition, the texture of the image is reconstructed while the edge is ensured, and the high-resolution image is obtained by solving the following equation:
wherein the content of the first and second substances,the method comprises the following steps of convolution operation, G represents a Gaussian kernel function, ↓ (n) represents downsampling with the proportion of n, Ψ is a set of gradient extraction operation, a first item of an equation is data fidelity, and a second item of the equation is a gradient constraint item.
In the image enlargement method of the present invention, to solve the equation, a gradient descent method is employed, as shown in the following formula (6):
wherein t is iteration times, τ is an iteration step size, ↓ (n) represents upsampling with a proportion of n, two Sobel operators in the horizontal direction and the vertical direction are adopted in gradient extraction, a Gaussian kernel function G is a Gaussian function with a mean value of 0 and a standard deviation of 1, a down-sampling proportion and an up-sampling proportion are both 2, the iteration step size τ is 0.1, and λ is 0.2. The number of iterations is 30.
Drawings
Fig. 1 is a main flowchart showing an image enlarging method of the present invention.
Fig. 2 is a sub-flowchart showing a gradient value estimation step of the image enlargement method of the present invention.
Fig. 3 is a sub-flowchart showing a high resolution gradient value estimation step of the image enlargement method of the present invention.
Fig. 4 is a diagram showing a spatial position relationship between a pixel to be interpolated and a known pixel.
Fig. 5 is a schematic diagram showing interpolation of white square-marked pixel points.
Fig. 6 is a schematic diagram showing interpolation of pixel points marked by white circles.
Fig. 7(a) to 7(b) are graphs showing comparison of effects before and after correction of an image.
Fig. 8(a) to 8(b) are comparison diagrams of a low-resolution image and a high-resolution image before and after reconstruction.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more clearly and completely understood, the technical solutions in the embodiments of the present invention will be described below with reference to the accompanying drawings in the embodiments of the present invention, and it should be understood that the specific embodiments described herein are only for explaining the present invention and are not intended to limit the present invention. The described embodiments are only some embodiments of the invention, not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in the flowchart of fig. 1, an image enlarging method of the present invention includes: a gradient value estimating step S1, wherein the gradient of the high-resolution image is estimated by adopting a modified non-local mean method; a gradient-guided interpolation step S2 in which interpolation of edge pixels is guided using gradient values; an interpolation result correction step S3, wherein a non-local mean filtering method used in a gradient estimation stage is adopted to carry out post-processing on the interpolation image, and noise and artificial effect introduced by interpolation are removed; and a texture reconstruction step S4, in which the interpolated image is used as an initial result, and its gradient is used as a constraint, and a reconstruction-based method is used to obtain a final high-resolution image.
As shown in the sub-flowchart of fig. 2, the gradient value estimating step S1 includes a similarity measure step S11 and a high resolution gradient estimating step S12. In the similarity measurement step S11, the similarity of the pixel points is defined by the similarity of the image blocks, assuming that the current pixel point is y (i, j), the image block formed by the peripheral N × N pixel points is N (i, j), and assuming that another pixel point in the image is y (m, N), the image block formed by the peripheral N × N pixel points is N (m, N), the similarity between the pixel points y (i, j) and y (m, N) is estimated by the gray level intensity similarity of the corresponding image blocks.
In addition, the gray-scale intensity difference between image blocks is defined by equation (1):
wherein the content of the first and second substances,is an operator of the second normal form,
a weight is given to the pixel point y (m, n) to measure the similarity, as shown in formula (2):
where Z (i, j) is a normalized constant representing the sum of all weights, and the parameter σ 1 controls the decay rate of the exponential equation. Here, the larger the difference between the image blocks is, the smaller the weight given to the corresponding pixel point is, and otherwise, the larger the weight is. The block size N × N is set to 7 × 7. The σ 1 size takes the variance of a 7 × 7 image block.
In addition, as shown in the sub-flowchart of fig. 3, in the high-resolution gradient estimation S12, the low-resolution image is first processed by using the conventional bicubic interpolation to obtain an initial high-resolution image, and then the initial high-resolution image is convolved by using the Sobel operator to obtain the gradient approximate estimation of the high-resolution image. Suppose that the low resolution image is enlarged twice, as shown in fig. 4, the black point is the position of the low resolution image in the high resolution image after enlargement, and the white point is the pixel point to be interpolated. And then, correcting the gradient of the pixel point to be interpolated. According to the similarity measurement of the pixel points in the previous step, y (i, j) is assumed to be any white pixel point to be interpolated in the image, and the initial gradient is G (i, j). y (m, n) is an arbitrary black known low resolution pixel point with a gradient of G (m, n). The gradient after y (i, j) correction is the weighted average of the gradients of all black points y (m, n) in the peripheral S multiplied by S window size, and the weight is the weight estimated in the step one. Here, the gradient modification is as shown in equation (3):
wherein the window sxs size is set to 21 × 21.
In addition, in the gradient-guided interpolation step, as shown in fig. 4, a white point is a pixel to be interpolated, and interpolation is performed in two steps. The first step is to interpolate the white square marked pixel points, and the second step is to interpolate the white round marked pixel points. The position relationship between the first-step interpolation pixel and the known pixel is shown in fig. 5, where Mi is a pixel to be interpolated, Nj, j is 1,2,3, and 4 are known low-resolution pixels, and the value of Mi is the weighted average of Nj. The plane is divided into 8 directional sections.
Subdividing the 8 directions into two categories, omega respectively1＝{β1,β3,β5,β7And Ω2＝{β0,β2,β4,β6}. Suppose the gradient direction of Mi is θMIf theta is greater than thetaMBelong to omega1As shown in fig. 5, the black point is projected to the gradient direction, and assuming that p (ij) is the projection length, the weight of Nj is defined as formula 4:
where c (i) is a normalization parameter representing the sum of all weights, σ 2 controls the decay rate of the exponential equation. If theta is greater than thetaMBelong to omega2As shown in fig. 5, the projection lengths of adjacent pixels are almost the same, that is, the weights are also almost the same, so that the weights need to be modified to maintain the sharpness of the edges in the direction. Method of correctionThe method is to compare the gradient strength of two sides of the pixel point to be interpolated, and if the gradient direction is the vertical direction, the gradient strength of the upper side and the lower side are compared, which are the sum of the gradients of the two points on the two sides respectively. If the difference between the gradients is larger than the threshold value T, the weight of the side with larger gradient is assigned as 0. If the gray level difference is smaller than T, the gray level difference between the pixels on the two sides and the pixel to be interpolated is continuously compared, and the gray levels of the pixels on the two sides are respectively the gray level average value of the two pixels on the two sides. The weight of the side with larger gray scale difference is assigned to 0. Where σ 2 is 0.2. The threshold T is taken to be 50. The position relationship between the interpolation pixel and the known pixel in the second step is shown in fig. 6. The interpolated pixels in the first step can be regarded as known pixels, the image is rotated by 45 degrees, and the spatial position relationship between the unknown pixel and the known pixel in the second step is consistent with that in fig. 5, so that the interpolation method in the first step is adopted.
The above interpolation introduces some jaggies while maintaining the edges, so the result of the interpolation is corrected in this step. As shown in fig. 4, for each pixel marked by a white point, the pixel marked by black in the peripheral sxs window is weighted and averaged to obtain a corrected gray value. And estimating the weight by using the similarity between the pixel points in the step one. A comparison of the results before and after correction is shown in FIG. 7. The window size is also 21 × 21.
The above steps are mainly to reconstruct the edge structure of the image. In the texture structure reconstruction step, the texture structure is reconstructed on the premise of ensuring the reconstructed edges. Assume that the input low resolution image is IlThe interpolated image isThe super-resolution technology based on reconstruction is adopted, the gradient of the interpolation image is used as a constraint condition, the texture of the image is reconstructed while the edge is ensured, and the high-resolution image is obtained by solving the following equation:
wherein the content of the first and second substances,the method comprises the following steps of convolution operation, G represents a Gaussian kernel function, ↓ (n) represents downsampling with the proportion of n, Ψ is a set of gradient extraction operation, a first item of an equation is data fidelity, and a second item of the equation is a gradient constraint item. λ controls the relative weight of the two parts, and the larger λ, the more the gradient constraint term weight, and the sharper edge can be generated. To solve the equation, a gradient descent method is used, as shown in the following equation (6):
wherein t is iteration times, τ is an iteration step size, ↓ (n) represents upsampling with a proportion of n, two Sobel operators in the horizontal direction and the vertical direction are adopted in gradient extraction, a Gaussian kernel function G is a Gaussian function with a mean value of 0 and a standard deviation of 1, a down-sampling proportion and an up-sampling proportion are both 2, the iteration step size τ is 0.1, and λ is 0.2. The number of iterations is 30. As shown in fig. 8, (a) is a low-resolution image before reconstruction, and (b) is a high-resolution image after reconstruction.
According to the image amplification method based on structure preservation, the local and non-local structure information of the image is fully utilized, in the gradient estimation, the characteristic that similar structures have similar gradient information is utilized, non-mean filtering based on gray intensity and gray distribution is adopted, the precision of the pixel to be interpolated is accurately estimated, and meanwhile, the algorithm is more robust. In the interpolation stage, the plane is divided into 8 direction intervals, and different interpolation methods are adopted for edges in different directions, so that the edge sharpness of each direction is maintained. After the interpolation is completed, the non-local mean filtering is adopted for post-processing, so that the aliasing effect introduced by the interpolation can be remarkably removed, and meanwhile, the sharpness of the edge is maintained. On the other hand, the image obtained by interpolation is used as an initial result, the gradient of the image is used as a constraint term, and the texture part of the image is reconstructed while the sharpness of the edge is kept by adopting a super-resolution technology based on reconstruction, so that the method can well reconstruct the edge and the texture of the image simultaneously, and lays a foundation for subsequent application.
The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention.
Claims (6)
1. A method of magnifying an image, characterized in that,
the method comprises the following steps:
a gradient value estimation step, wherein the gradient of the high-resolution image is estimated by adopting a non-local mean filtering method;
a gradient guiding interpolation step, wherein the gradient value is used for guiding the interpolation of the edge pixel;
an interpolation result correction step, wherein the interpolation image is post-processed by adopting a non-local mean method used in the gradient estimation step, and noise and artificial effect introduced by interpolation are removed; and
a texture structure reconstruction step, in which the interpolated image is used as an initial result, the gradient thereof is used as a constraint, and the image is reconstructed to obtain a final high-resolution image,
the gradient value estimating step includes a similarity measuring step and a high resolution gradient estimating step,
in the similarity measurement step, the similarity of the pixel points is defined through the similarity of the image blocks, wherein, the current pixel point is assumed to be y (i, j), the image block formed by the peripheral N multiplied by N pixel points is assumed to be N (i, j), another pixel point in the image is assumed to be y (m, N), the image block formed by the peripheral N multiplied by N pixel points is assumed to be N (m, N), the similarity between the pixel points y (i, j) and y (m, N) is estimated through the gray intensity similarity of the corresponding image blocks,
the difference in gray scale intensity between the image blocks is defined by equation (1):
wherein the content of the first and second substances,is an operator of the second normal form,
a weight is given to the pixel point y (m, n) to measure the similarity, as shown in formula (2):
wherein Z (i, j) is a normalized constant representing the sum of all weights, the parameter σ 1 controls the decay rate of the exponential equation,
in the high-resolution gradient estimation step, a low-resolution image is processed by using the traditional bicubic interpolation to obtain an initial high-resolution image, then the initial high-resolution image is subjected to convolution operation by using a Sobel operator to obtain the gradient approximate estimation of the high-resolution image, and the gradient of a pixel point to be interpolated is corrected.
2. The image enlarging method according to claim 1,
the gradient modification is shown as formula (3):
wherein the window sxs size is set to 21 × 21.
3. The image enlarging method according to claim 1 or 2,
in the step of gradient guiding interpolation, a plane is divided into 8 direction intervals,
subdividing the 8 directions into two categories, omega respectively1＝{β1,β3,β5,β7And Ω2＝{β0,β2,β4,β6Suppose the gradient direction of Mi is θMIf theta is greater than thetaMBelong to omega1Projecting the black point to the gradient direction, and assuming p (ij) as the projection length, the weight of Nj is defined as formula 4:
where c (i) is a normalization parameter representing the sum of all weights, σ 2 controls the decay rate of the exponential equation.
4. The image enlarging method according to claim 3,
and assigning the weight value of the side with larger gray difference as 0, wherein sigma 2 is 0.2, and the threshold value T is 50.
5. The image enlarging method according to claim 1,
in the texture reconstruction, the input low resolution image is assumed to be IlThe interpolated image isThe super-resolution technology based on reconstruction is adopted, the gradient of the interpolation image is used as a constraint condition, the texture of the image is reconstructed while the edge is ensured, and the high-resolution image is obtained by solving the following equation:
wherein the content of the first and second substances,is convolution operation, G represents Gaussian kernel function, ↓ (n) represents downsampling with proportion of n, Ψ is a set of gradient extraction operation, the first item of the equation is data fidelity, the second itemIs a gradient constraint term.
6. The image enlarging method according to claim 5,
to solve the equation, a gradient descent method is used, as shown in the following equation (6):
wherein t is iteration frequency, τ is iteration step size, ↓ (n) represents upsampling with the proportion of n, two Sobel operators in the horizontal direction and the vertical direction are adopted in gradient extraction, a Gaussian kernel function G is a Gaussian function with the mean value of 0 and the standard deviation of 1, the down-sampling proportion and the up-sampling proportion are both 2, the iteration step size τ is 0.1, λ is 0.2, and the iteration frequency is 30.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610509556.2A CN106169174B (en) | 2016-06-30 | 2016-06-30 | Image amplification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610509556.2A CN106169174B (en) | 2016-06-30 | 2016-06-30 | Image amplification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106169174A CN106169174A (en) | 2016-11-30 |
| CN106169174B true CN106169174B (en) | 2019-12-31 |
Family
ID=58065970
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610509556.2A Active CN106169174B (en) | 2016-06-30 | 2016-06-30 | Image amplification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106169174B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106910215B (en) * | 2017-03-15 | 2022-01-07 | 沈阳理工大学 | Super-resolution method based on fractional order gradient interpolation |
| CN107194877A (en) * | 2017-06-16 | 2017-09-22 | 南京大学金陵学院 | A kind of guarantor side interpolation super-resolution computational methods based on single image |
| CN107316323B (en) * | 2017-06-28 | 2020-09-25 | 北京工业大学 | No-reference image quality evaluation method established based on multi-scale analysis method |
| CN107967669B (en) * | 2017-11-24 | 2022-08-09 | 腾讯科技（深圳）有限公司 | Picture processing method and device, computer equipment and storage medium |
| CN110763342B (en) * | 2019-09-30 | 2020-12-22 | 成都鼎屹信息技术有限公司 | Method for restoring resolution of infrared polarization super-pixel radiation intensity image |
| CN116912096B (en) * | 2023-09-14 | 2024-01-09 | 惠州市耀盈精密技术有限公司 | IMOLD system-based precision mold template image generation method |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103177420A (en) * | 2013-03-13 | 2013-06-26 | 北京大学 | Image amplification method and image application device based on local-area feature correlations |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2014083857A1 (en) * | 2012-11-29 | 2014-06-05 | 日本電気株式会社 | Image processing device and image processing method |
-
2016
- 2016-06-30 CN CN201610509556.2A patent/CN106169174B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103177420A (en) * | 2013-03-13 | 2013-06-26 | 北京大学 | Image amplification method and image application device based on local-area feature correlations |
Non-Patent Citations (2)
| Title |
|---|
| Image Super-Resolution Based on Structure-Modulated Sparse Representation;Yongqin Zhang等;《IEEE Transactions on Image Processing》;20150508;第24卷(第9期);第2797-2810页 * |
| 改进的图像自适应梯度插值;张美玉 等;《中国图象图形学报》;20090531;第14卷(第5期);第853-858页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106169174A (en) | 2016-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106169174B (en) | Image amplification method | |
| Zhao et al. | Fast Single Image Super-Resolution Using a New Analytical Solution for $\ell _ {2} $–$\ell _ {2} $ Problems | |
| Kamilov | A parallel proximal algorithm for anisotropic total variation minimization | |
| Yang et al. | Single-image super-resolution reconstruction via learned geometric dictionaries and clustered sparse coding | |
| US8731337B2 (en) | Denoising and artifact removal in image upscaling | |
| CN106169173B (en) | Image interpolation method | |
| US9619755B2 (en) | Method for estimating parameters of a graph spectral filter using training data | |
| CN107133923B (en) | Fuzzy image non-blind deblurring method based on adaptive gradient sparse model | |
| CN110796622B (en) | Image bit enhancement method based on multi-layer characteristics of series neural network | |
| CN110675347A (en) | Image blind restoration method based on group sparse representation | |
| Mahajan et al. | Adaptive and non-adaptive image interpolation techniques | |
| Nasonov et al. | Fast super-resolution using weighted median filtering | |
| CN109741258B (en) | Image super-resolution method based on reconstruction | |
| Chen et al. | Bayesian video super-resolution with heavy-tailed prior models | |
| CN112837220B (en) | Method for improving resolution of infrared image and application thereof | |
| CN113963009A (en) | Local self-attention image processing method and model based on deformable blocks | |
| US20240054605A1 (en) | Methods and systems for wavelet domain-based normalizing flow super-resolution image reconstruction | |
| KR101341617B1 (en) | Apparatus and method for super-resolution based on error model of single image | |
| CN110136061B (en) | Resolution improving method and system based on depth convolution prediction and interpolation | |
| CN107767342B (en) | Wavelet transform super-resolution image reconstruction method based on integral adjustment model | |
| Fan et al. | Single image super resolution method based on edge preservation | |
| Zhao et al. | Fast single image super-resolution | |
| CN113506212A (en) | Improved POCS-based hyperspectral image super-resolution reconstruction method | |
| Liu et al. | 32‐2: Improvement in Directional Cubic Convolution Image Interpolation | |
| Xiao et al. | A high-efficiency edge-preserving Bayesian method for image interpolation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20161130 Assignee: Boya cloud (Beijing) Technology Co., Ltd. Assignor: Peking University Contract record no.: 2017990000367 Denomination of invention: Image amplification method and device License type: Common License Record date: 20170908 |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |