CN116503249A - Real world image super-resolution dataset construction method and system - Google Patents
Real world image super-resolution dataset construction method and system Download PDFInfo
- Publication number
- CN116503249A CN116503249A CN202310393377.7A CN202310393377A CN116503249A CN 116503249 A CN116503249 A CN 116503249A CN 202310393377 A CN202310393377 A CN 202310393377A CN 116503249 A CN116503249 A CN 116503249A
- Authority
- CN
- China
- Prior art keywords
- resolution
- resolution image
- image
- low
- raw
- 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.)
- Pending
Links
- 238000010276 construction Methods 0.000 title claims abstract description 8
- 238000000034 method Methods 0.000 claims abstract description 43
- 238000012937 correction Methods 0.000 claims abstract description 40
- 238000010586 diagram Methods 0.000 claims description 31
- 239000011159 matrix material Substances 0.000 claims description 23
- 238000005070 sampling Methods 0.000 claims description 4
- 230000000295 complement effect Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 8
- 238000004088 simulation Methods 0.000 abstract description 3
- 238000006731 degradation reaction Methods 0.000 description 6
- 230000015556 catabolic process Effects 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 239000011248 coating agent Substances 0.000 description 2
- 238000000576 coating method Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000004026 adhesive bonding Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 230000003287 optical effect Effects 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
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- 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
- G06T3/4015—Image demosaicing, e.g. colour filter arrays [CFA] or Bayer patterns
-
- 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
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
- G06T3/4076—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution using the original low-resolution images to iteratively correct the high-resolution images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/90—Dynamic range modification of images or parts thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/90—Determination of colour characteristics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a method for constructing a super-resolution dataset of a real world image, which comprises the following steps: acquiring a scene high-resolution image, and acquiring a same scene low-resolution image; carrying out RAW domain local color and brightness correction on the high-resolution image and the low-resolution image; carrying out RAW domain histogram matching on the high-resolution image and the low-resolution image; carrying out RAW domain alignment on the high-resolution image and the low-resolution image to obtain a RAW image; demosaicing is carried out on the RAW graph to obtain an RGB graph; gamma correction is carried out on the RGB map, and a real world RGB domain data set is obtained; meanwhile, the invention also discloses a real world image super-resolution data set construction system using the method, which comprises the following steps: platform, spectroscope, high resolution camera and low resolution camera. The invention solves the problem that the network trained on the simulation data set has poor effect when applied to the picture in the real scene, so as to be suitable for the real world.
Description
Technical Field
The invention relates to the technical field of computer vision graphics, in particular to a method and a system for constructing a super-resolution dataset of a real-world image.
Background
Super resolution (Single Image Super Resolution, SISR) of a single image is a very widely applied technology, and in the information age, the picture is very frequently enlarged, so SISR is a computer vision technology in research hotspots for a long time. In the past few years, with the development of deep neural network technology, the performance of the traditional image super-resolution technology is gradually surpassed by the image super-resolution method based on deep learning. A significant feature of deep learning is data dependence, and most image super-resolution methods also require paired high-low resolution picture pair datasets at present, with the most commonly used datasets being DIV2K and Flickr2K. The low resolution pictures of the two data sets are obtained by degradation operations such as blurring and downsampling of the high resolution pictures, so that such data sets can be classified as analog data sets compared to the real photographed paired data sets.
Recently, real world super-resolution topics are becoming the focus of research. The image super-resolution method based on deep learning has better performance on the analog data set, however, when the network trained on the analog data set is applied to the picture in the real scene, the network tends to have poorer effect, and the essential reason is that the degradation process of the real world is more complex and far from being covered by the degradation space of the analog data set.
Disclosure of Invention
The invention aims to provide a method and a system for constructing a super-resolution data set of a real world image, which solve the problem that a network trained on an analog data set has poor effect when applied to a picture in a real scene, so as to be suitable for the real world.
The embodiment of the invention is realized by the following technical scheme:
in one aspect, the invention provides a method for constructing a super-resolution dataset of a real world image, comprising the following steps:
acquiring a scene high-resolution image, and acquiring a same scene low-resolution image;
carrying out RAW domain local color and brightness correction on the high-resolution image and the low-resolution image;
carrying out RAW domain histogram matching on the high-resolution image and the low-resolution image;
carrying out RAW domain alignment on the high-resolution image and the low-resolution image to obtain a RAW image;
demosaicing is carried out on the RAW graph to obtain an RGB graph;
gamma correction is carried out on the RGB map, and a real world RGB domain data set is obtained.
In an embodiment of the present invention, the specific method for acquiring the high resolution image of the scene and then acquiring the low resolution image of the same scene is as follows:
the spectroscope is adopted, and a high-resolution camera and a low-resolution camera are respectively erected above and at one side of the spectroscope, so that a high-resolution image and a low-resolution image of the same scene are obtained.
In an embodiment of the present invention, the specific method for performing RAW domain local color and brightness correction on the high resolution image and the low resolution image is as follows:
obtaining calibration RAW diagram I 1 And calibrating RAW diagram I 2 Alignment calibration RAW diagram I 1 And calibrating RAW diagram I 2 Performing a local gray world hypothesis white balance algorithm on the corresponding region to obtain a calibration matrix, and applying the calibration matrix to color correction of the high-resolution image and the low-resolution image;
and acquiring Bayer areas of the high-resolution image and the low-resolution image, solving the average value of the two Bayer areas, obtaining the average value ratio as a brightness correction coefficient, and applying the brightness correction coefficient to brightness correction of the low-resolution image.
In one embodiment of the present invention, the step of obtaining a calibration RAW chart I 1 And calibrating RAW diagram I 2 The method of (2) is as follows:
acquiring calibration RAW image I by using high-resolution camera to shoot gray card 1 By low scoreRaw image I for resolution camera shooting calibration 2 。
In an embodiment of the present invention, the specific method for obtaining the calibration matrix is as follows:
in the calibration of RAW diagram I 1 And calibrating RAW diagram I 2 The corresponding Bayer regions of 64x64 and 32x32 are divided to obtain R, G according to the formula 1 、G 2 Correction coefficient k of four channels B r 、k g1 、k g2 、k b And combining correction coefficients of all areas to obtain a calibration matrix.
In an embodiment of the present invention, a specific method for obtaining the brightness correction coefficient is as follows:
the high resolution image is divided into Bayer areas of 64x64, the low resolution image is divided into Bayer areas of 32x32, the average value of the two Bayer areas is obtained, and the average value ratio is the brightness correction coefficient.
In an embodiment of the present invention, the specific method for performing RAW domain histogram matching on a high resolution image and a low resolution image is as follows:
dividing RG in RAW domain 1 G 2 The B channel performs global color calibration on the high resolution image and the low resolution image.
In an embodiment of the present invention, the specific method for performing RAW domain alignment on a high resolution image and a low resolution image is as follows:
calculating in an RGB domain to obtain a homography matrix, and applying the homography matrix to a low-resolution image to obtain a channel diagram after spatial warping;
and (3) carrying out nearest neighbor downsampling on each channel map, and carrying out the inverse process of channel extraction to obtain the finished registration.
In an embodiment of the present invention, the specific method for obtaining the homography matrix is as follows:
extracting R, G the high resolution image and the low resolution image, respectively 1 G 2 The corresponding R, G is obtained by carrying out nearest neighbor up-sampling and complement on the three channels B to form a complete image 1 G 2 B, a channel diagram;
performing bicubic downsampling on each channel map of the high-resolution image to obtain HR-RAW-bic;
and detecting the features by a SIFT algorithm through the channel diagrams of the HR-RAW-bic and the low-resolution image to obtain feature descriptors, matching the extracted features by a RANSAC algorithm, and minimizing errors among all the corresponding features to obtain a homography matrix after feature fitting.
On the other hand, the invention also provides a real world image super-resolution data set construction system using the method, which comprises the following steps:
the platform is provided with two cloud platforms;
the spectroscope is arranged in the platform;
the high-resolution camera is installed on one of the cloud platforms;
and the low-resolution camera is mounted on the other cloud deck.
The technical scheme of the embodiment of the invention has at least the following advantages and beneficial effects:
the invention provides a real-world image super-resolution dataset construction method, which is used for carrying out RAW domain local color, brightness correction, RAW domain histogram matching, RAW domain alignment, demosaicing processing and gamma correction on a high-resolution image and a low-resolution image of a real world, so that the problem that the effect of a network trained on a simulation dataset is poor when the network is applied to a picture in the real scene is solved, and the real-world image super-resolution dataset construction method is better suitable for the real world.
Drawings
FIG. 1 is a flow chart of the present invention;
FIG. 2 is a schematic diagram of an imaging system of the present invention;
FIG. 3 is an imaging schematic of the present invention;
FIG. 4 is a diagram showing the effect of data set processing;
FIG. 5 is a superdivision effect display diagram;
fig. 6 is a view showing the effect of cross-camera generalization experiments.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
Referring to fig. 1, a method for constructing a super-resolution dataset of a real world image includes the following steps:
101. acquiring a scene high-resolution image, and acquiring a same scene low-resolution image;
102. carrying out RAW domain local color and brightness correction on the high-resolution image and the low-resolution image;
103. carrying out RAW domain histogram matching on the high-resolution image and the low-resolution image;
104. carrying out RAW domain alignment on the high-resolution image and the low-resolution image to obtain a RAW image;
105. demosaicing is carried out on the RAW graph to obtain an RGB graph;
106. gamma correction is carried out on the RGB map, and a real world RGB domain data set is obtained.
In step 101, as shown in fig. 2, the viewing angle positions of the high-resolution camera and the low-resolution camera are aligned, a beam splitter is installed, and light enters from the beam splitter and is split into reflected light and refracted light to enter the high-resolution camera and the low-resolution camera respectively. As shown in fig. 3, under the condition of ensuring that the object distance d and the focal length f are unchanged, the number of pixels contained in the camera sensor is different due to the difference of the pixel sizes of the sensor, so that a high-resolution image HR (HR-RAW) and a low-resolution image LR (LR-RAW) of the same scene, namely high-low resolution paired pictures, can be simultaneously shot.
In this embodiment, parameters of the high-resolution camera and the low-resolution camera are as follows:
it should be noted that, the high resolution pictures HR and LR directly shot by the high resolution camera and the low resolution camera cannot be directly used for the training data of the super resolution network, and the color and brightness of the HR and LR pictures are slightly different from each other due to the spectroscope and the sensor pixel error of the most original high-low resolution paired RAW pictures, so that the data are processed in steps 102-106, so that the color and space error between the HR and LR pictures is as small as possible.
Specifically, in step 102, the specific method of color correction is as follows: acquiring calibration RAW map I by photographing gray cards with high resolution camera and low resolution camera 1 And calibrating RAW diagram I 2 Alignment calibration RAW diagram I 1 And calibrating RAW diagram I 2 Performing a local gray world hypothesis white balance algorithm on the corresponding region to obtain a calibration matrix, and applying the calibration matrix to color correction of the high-resolution image and the low-resolution image;
the acquisition method of the calibration matrix comprises the following steps: in the calibration of RAW diagram I 1 And calibrating RAW diagram I 2 The corresponding Bayer areas divided into 64x64 and 32x32, each of which conforms to the gray world hypothesis, have the formula:
wherein,,mean value of red pixels in Bayer region, < >>Refers to the average value of the Bayer region. According to the above, R, G 1 、G 2 Correction coefficient k of four channels B r 、k g1 、k g2 、k b . The correction coefficients of all areas are combined to obtain the correction matrices M1 and M2, and the correction of colors can be completed by respectively applying the correction matrices to the photographed HR-RAW and LR-RAW.
In step 102, the specific method for brightness correction is: dividing the HR-RAW and the LR-RAW into Bayer areas of 64x64 and 32x32, averaging each area, wherein the average value ratio of the two areas is the brightness correction coefficient, and multiplying the brightness correction coefficients of all areas with the corresponding areas of the LR-RAW to finish the brightness correction.
After step 102 is completed, since the HR and LR pictures may still have color differences as a whole after the primary calibration of color and brightness is completed, step 103 is also required to perform RAW domain histogram matching on the HR and LR pictures, that is, in RAW domain RG 1 G 2 The B channel further performs global color calibration.
After step 103 is finished, performing step 104 to perform RAW domain alignment on the HR picture and the LR picture, firstly calculating in the RGB domain to obtain a homography matrix H, and then applying the homography matrix H to LR-RAW to complete registration to obtain a RAW image, namely a real world RAW domain data set RAW BS-realSR;
specifically, HR-RAW and LR-RAW are first extracted R, G respectively 1 G 2 The corresponding R, G is obtained by carrying out nearest neighbor up-sampling and complement on the three channels B to form a complete image 1 G 2 B, a channel diagram; then each channel diagram of the HR-RAW is subjected to bicubic downsampling to obtain HR-RAW-bic; performing SIFT algorithm on the channel diagrams of HR-RAW-bic and LR-RAW to detect the features to obtain feature descriptors, matching the extracted features by RANSAC algorithm, and minimizing the error between all the corresponding features to obtain the feature-fitted featuresHomography matrix H.
Multiplying each channel map of LR-RAW with homography matrix H to obtain channel map after spatial distortion, sampling down in nearest neighbor, and obtaining registered HR-RAW and LR-RAW image pair after channel extraction inverse process, namely obtaining RAW map.
In step 105, demosaicing is performed on the RAW image obtained in step 104, so as to obtain an RGB image. And step 106, gamma correction is performed on the RGB map, and the brightness of the picture is raised, so that a real world RGB domain data set RGB BS-realSR is finally obtained, as shown in FIG. 4.
Example 2
In order to compare the proposed dataset BS-RealSR with the simulated dataset, a total of three datasets were used for the experiment, respectively the most commonly used DIV2K for image super resolution, BS-RealSR and BS-RealSR HR degradation proposed in this example generated BS-RealSR-syn for the corresponding LR picture.
Training on the three data sets by using the currently mainstream super-resolution networks EDSR and RCAN, testing on a real-world data set BS-RealSR-testset, and finally evaluating the network super-resolution effect by using PSNR, SSIM and LPIPS evaluation indexes to obtain the following table.
From the experimental results and fig. 5, BS-RealSR has better performance in the real world application scenario, and BS-RealSR-syn and DIV2K perform nearly, but are not as similar to BS-RealSR, which indicates that the data set generated by simulation hardly covers the complex and diverse degradation conditions of the real world.
Example 3
This example performs a cross-camera experiment to illustrate the generalization of the BS-RealSR dataset. Networks trained on different data sets were tested on a RealSR, which is a high-low resolution paired data set taken through a zoom lens using two cameras, cannon 5D3 and Nikon D810.
Experimental results and fig. 6 it can be seen that the network trained on BS-RealSR performed somewhat less well in the face of the RealSR test set, but still performed better than the two comparative simulated data sets. This illustrates that the BS-RealSR dataset, although not photographed by the same camera as the RealSR dataset, has far less spatial differences in degradation than between the analog dataset and the real dataset.
Example 4
A real world image super-resolution dataset construction system using the method of embodiments 1-3, as shown in fig. 2-3, comprising a platform, a spectroscope, a high-resolution camera and a low-resolution camera, the spectroscope being disposed in the platform, the platform being provided with two holders, the high-resolution camera and the low-resolution camera being mounted respectively.
The spectroscope splits incident light rays of a shooting scene into incident light rays and reflected light rays, and the two light rays enter the high-resolution camera and the low-resolution camera respectively. The lenses of the high-resolution camera and the low-resolution camera are identical, except that the pixel size of the sensor film is different, so that in the case of identical sensor film areas, when the pixel size of the low-resolution camera is 2 times that of the high-resolution camera, it means that the resolution of the picture taken by the high-resolution camera is 2 times that of the low-resolution camera.
In this embodiment, the beam splitter is an optical component capable of dividing incident light into two beams of reflected light and transmitted light according to different ratios, and is generally formed by gluing two prisms, wherein the hypotenuse of one prism indicates that a coating is performed, and the specific application of the beam splitter prism is determined by the nature of the coating. The present embodiment employs a 50:50 neutral beam splitting prism.
In this embodiment, the pan-tilt has four axises, so that the high-resolution camera and the low-resolution camera can be finely tuned in four directions of front, back, left and right on the horizontal plane, and can be finely tuned in the vertical direction up and down, and can be rotationally finely tuned.
The working principle of the embodiment is as follows:
the high-resolution camera and the low-resolution camera are respectively carried on the four-axis cradle head, and the centers of the images shot by the high-resolution camera and the low-resolution camera are aligned after the positions of the high-resolution camera and the low-resolution camera are finely adjusted; the high-resolution camera and the low-resolution camera are closely attached to the beam splitter prism, and respectively capture the refraction light and the transmission light of the beam splitter.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. The method for constructing the super-resolution data set of the real world image is characterized by comprising the following steps of:
acquiring a scene high-resolution image, and acquiring a same scene low-resolution image;
carrying out RAW domain local color and brightness correction on the high-resolution image and the low-resolution image;
carrying out RAW domain histogram matching on the high-resolution image and the low-resolution image;
carrying out RAW domain alignment on the high-resolution image and the low-resolution image to obtain a RAW image;
demosaicing is carried out on the RAW graph to obtain an RGB graph;
gamma correction is carried out on the RGB map, and a real world RGB domain data set is obtained.
2. The method for constructing a super-resolution dataset of a real-world image according to claim 1, wherein the specific method for acquiring the high-resolution image of the scene and then acquiring the low-resolution image of the same scene is as follows:
the spectroscope is adopted, and a high-resolution camera and a low-resolution camera are respectively erected above and at one side of the spectroscope, so that a high-resolution image and a low-resolution image of the same scene are obtained.
3. The method for constructing super-resolution dataset of real-world image according to claim 1, wherein the specific method for performing RAW domain local color and brightness correction for high-resolution image and low-resolution image is as follows:
obtaining calibration RAW diagram I 1 And calibrating RAW diagram I 2 Alignment calibration RAW diagram I 1 And calibrating RAW diagram I 2 Performing a local gray world hypothesis white balance algorithm on the corresponding region to obtain a calibration matrix, and applying the calibration matrix to color correction of the high-resolution image and the low-resolution image;
and acquiring Bayer areas of the high-resolution image and the low-resolution image, solving the average value of the two Bayer areas, obtaining the average value ratio as a brightness correction coefficient, and applying the brightness correction coefficient to brightness correction of the low-resolution image.
4. The method of claim 3, wherein the step of obtaining a calibration RAW map I 1 And calibrating RAW diagram I 2 The method of (2) is as follows:
acquiring calibration RAW image I by using high-resolution camera to shoot gray card 1 Calibrating RAW map I using low resolution camera shooting 2 。
5. A method of constructing a super-resolution dataset of a real world image as claimed in claim 3, wherein the specific method of obtaining the calibration matrix is as follows:
in the calibration of RAW diagram I 1 And calibrating RAW diagram I 2 The corresponding Bayer regions of 64x64 and 32x32 are divided to obtain R, G according to the formula 1 、G 2 Correction coefficient k of four channels B r 、k g1 、k g2 、k b And combining correction coefficients of all areas to obtain a calibration matrix.
6. A method of constructing a super-resolution dataset of a real world image as claimed in claim 3, wherein the specific method of obtaining the luminance correction factor is as follows:
the high resolution image is divided into Bayer areas of 64x64, the low resolution image is divided into Bayer areas of 32x32, the average value of the two Bayer areas is obtained, and the average value ratio is the brightness correction coefficient.
7. The method for constructing a super-resolution dataset of a real-world image according to claim 1, wherein the specific method for performing RAW domain histogram matching on the high-resolution image and the low-resolution image is as follows:
dividing RG in RAW domain 1 G 2 The B channel performs global color calibration on the high resolution image and the low resolution image.
8. The method for constructing a super-resolution dataset of a real world image according to claim 1, wherein the specific method for performing RAW domain alignment on the high resolution image and the low resolution image is as follows:
calculating in an RGB domain to obtain a homography matrix, and applying the homography matrix to a low-resolution image to obtain a channel diagram after spatial warping;
and (3) carrying out nearest neighbor downsampling on each channel map, and carrying out the inverse process of channel extraction to obtain the finished registration.
9. The method for constructing a super-resolution dataset of a real world image as claimed in claim 8, wherein the specific method for obtaining the homography matrix is as follows:
extracting R, G the high resolution image and the low resolution image, respectively 1 G 2 The corresponding R, G is obtained by carrying out nearest neighbor up-sampling and complement on the three channels B to form a complete image 1 G 2 B, a channel diagram;
performing bicubic downsampling on each channel map of the high-resolution image to obtain HR-RAW-bic;
and detecting the features by a SIFT algorithm through the channel diagrams of the HR-RAW-bic and the low-resolution image to obtain feature descriptors, matching the extracted features by a RANSAC algorithm, and minimizing errors among all the corresponding features to obtain a homography matrix after feature fitting.
10. A real world image super-resolution dataset construction system using the method of any of claims 1-9, comprising:
the platform is provided with two cloud platforms;
the spectroscope is arranged in the platform;
the high-resolution camera is installed on one of the cloud platforms;
and the low-resolution camera is mounted on the other cloud deck.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310393377.7A CN116503249A (en) | 2023-04-13 | 2023-04-13 | Real world image super-resolution dataset construction method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310393377.7A CN116503249A (en) | 2023-04-13 | 2023-04-13 | Real world image super-resolution dataset construction method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116503249A true CN116503249A (en) | 2023-07-28 |
Family
ID=87317581
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310393377.7A Pending CN116503249A (en) | 2023-04-13 | 2023-04-13 | Real world image super-resolution dataset construction method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116503249A (en) |
-
2023
- 2023-04-13 CN CN202310393377.7A patent/CN116503249A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Abdelhamed et al. | A high-quality denoising dataset for smartphone cameras | |
US10897609B2 (en) | Systems and methods for multiscopic noise reduction and high-dynamic range | |
CN111353948B (en) | Image noise reduction method, device and equipment | |
CN108776980A (en) | A kind of scaling method towards lenticule light-field camera | |
CN102844788A (en) | Image processing apparatus and image pickup apparatus using the same | |
CN109493283A (en) | A kind of method that high dynamic range images ghost is eliminated | |
CN113676628A (en) | Multispectral sensor, imaging device and image processing method | |
CN113676629B (en) | Image sensor, image acquisition device, image processing method and image processor | |
CN110519585A (en) | A kind of imaging calibration method and device applied to image capture device | |
CN102158731A (en) | Image processing system and method | |
CN113506275B (en) | Urban image processing method based on panorama | |
JPH1069543A (en) | Method and device for reconstituting curved surface of object | |
CN114463196A (en) | Image correction method based on deep learning | |
KR102437510B1 (en) | Plenoptic sub-aperture view shuffling for richer color sampling | |
KR102314719B1 (en) | Plenoptic sub-aperture view shuffling with improved resolution | |
CN111866369B (en) | Image processing method and device | |
CN116503249A (en) | Real world image super-resolution dataset construction method and system | |
CN102158730A (en) | Image processing system and method | |
Bigué et al. | Production of high-resolution reference polarization images from real world scenes | |
Sun et al. | Blind calibration for focused plenoptic cameras | |
CN108230273A (en) | A kind of artificial compound eye camera three dimensional image processing method based on geological information | |
Szeliski et al. | Image formation | |
LU102214B1 (en) | Method for digital image processing | |
US20230419446A1 (en) | Method for digital image processing | |
Volkova et al. | Software compensation of chromatic-aberration effects on color photographs |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |