CN105678728A - High-efficiency super-resolution imaging device and method with regional management - Google Patents
High-efficiency super-resolution imaging device and method with regional management Download PDFInfo
- Publication number
- CN105678728A CN105678728A CN201610033903.9A CN201610033903A CN105678728A CN 105678728 A CN105678728 A CN 105678728A CN 201610033903 A CN201610033903 A CN 201610033903A CN 105678728 A CN105678728 A CN 105678728A
- Authority
- CN
- China
- Prior art keywords
- resolution
- image
- super
- low
- dictionary
- 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
- 238000003384 imaging method Methods 0.000 title claims abstract description 65
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000012549 training Methods 0.000 claims abstract description 58
- 230000003287 optical effect Effects 0.000 claims abstract description 11
- 238000007781 pre-processing Methods 0.000 claims abstract description 11
- 239000011159 matrix material Substances 0.000 claims description 13
- 230000015572 biosynthetic process Effects 0.000 claims description 9
- 238000001914 filtration Methods 0.000 claims description 8
- 239000000284 extract Substances 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 238000003708 edge detection Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 238000002059 diagnostic imaging Methods 0.000 abstract description 3
- 238000012544 monitoring process Methods 0.000 abstract description 2
- 238000005516 engineering process Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 4
- 238000012937 correction Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000002146 bilateral effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000012804 iterative process Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- 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
- G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- 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
- G06T2207/10004—Still image; Photographic image
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention provides a high-efficiency super-resolution imaging device and method with regional management, and aims at solving the problem that an existing super-resolution imaging device and method is long in imaging time and low in resolution. The imaging device comprises an imaging lens group, a high-resolution detector, an image storage module, an image pre-processing module, an image super-resolution reconstruction module and an image output display module connected successively, wherein the image super-resolution reconstruction module is composed of an image regional management sub-module and a dictionary training and regional reconstruction sub-module. The imaging method comprises the following steps of obtaining optical signals of a practical scene; obtaining low-resolution images; storing the low-resolution images; preprocessing the images; carrying out regional image management; training dictionaries and reconstructing super-resolution areas, and splicing images of the reconstructed super-resolution sub-regions. The imaging device and method can be used to improve the imaging resolution effectively and shorten the imaging time, and applied to the fields including video monitoring, satellite remote-sensing imaging and medical imaging.
Description
Technical field
The invention belongs to Image Reconstruction Technology field, be specifically related to the efficient super-resolution imaging device and method of a kind of picture portion territory management, it is possible to for fields such as video monitoring, satellite remote sensing imaging, medical imagings.
Background technology
Along with developing rapidly of the technology such as high definition digital television, satellite remote sensing imaging, medical imaging, high-resolution imaging is required more and more higher by national economy social development. Improving resolution most straightforward approach is exactly increase the focal length of optical system or reduce CCD pixel unit size. But, increase system focal and can increase the volume of imaging system, improve cost; Reduce CCD pixel size and can introduce picture noise etc., therefore, the method that integrated software processes becomes research focus, the super-resolution rebuilding technology adopting image can not increase on the basis of complex hardware equipment, increased the detail of the high frequency of image by the image of low resolution by the method for image procossing, produce single width high-quality, high-resolution image.
Current super-resolution imaging device includes imaging lens group, high-resolution detector, image storage module, image super-resolution rebuilding module and image output display module five major part, the existing method wherein adopted in image super-resolution rebuilding module has multiframe super-resolution rebuilding and single frames super-resolution rebuilding two kinds, restriction due to real-time, single frames super-resolution imaging technology more practicality, it mainly has based on rebuilding and based on two kinds of methods of study, higher based on the method efficiency rebuild, but due to image blurring, noise, the impact of the factors such as displacement, reconstructed image quality can decline to some extent. method based on study effectively solves this problem, and its principle is to first pass through the relation between training low-resolution image and high-definition picture, is then reconstructed a panel height image in different resolution by the low-resolution image inputted in learning process.
The imaging device of existing super-resolution image and method, the Chinese patent application of such as Beijing Z-Good Technology Service Co., Ltd., application publication number is CN104320596A, name is called in the invention of " acquisition methods of super-resolution image and acquisition device ", propose a kind of pixel distribution by regulating the imageing sensor of each camera in camera array and obtain scene image, again the scene image obtained is rebuild, the required precision when command displacement of this device is higher, technical difficulty is big, the at least local that cost is high and super-resolution rebuilding region is photographed scene.Super resolution ratio reconstruction method based on study, the Chinese patent application of such as Peking University, application publication number is CN104103052A, name is called in the invention of " super resolution ratio reconstruction method based on study ", propose sample image is divided into notable district and non-significant district and is respectively trained dictionary, by solving non-significant sparse coefficient and being multiplied with dictionary and obtain high-definition picture block, but it is because the method, in calculating process, view picture image to be reconstructed is done same treatment simultaneously, so large scale image processing efficiency to be reconstructed is relatively low, and memory consumption is bigger, simultaneously because algorithm reason, cause that rebuilding image can exist serious puppet reconstruction part, affect reconstructed image quality. for another example the Chinese patent application of Xian Electronics Science and Technology University, application publication number is CN103295197A, name is called in the invention of " the Image Super-resolution Reconstruction method based on dictionary learning and bilateral canonical ", proposition extracts high-frequency characteristic after image to be reconstructed does different scale interpolation, characteristic is carried out dictionary training, but due to the limitation of its dictionary training sample, thus reconstructed image resolution is also affected by constraint.
Summary of the invention
It is an object of the invention to the defect overcoming above-mentioned prior art to exist, it is proposed that the efficient super-resolution imaging device and method of a kind of district management, for solving existing super-resolution imaging device and method imaging time length and the low technical problem of resolution.
To achieve these goals, the technical scheme that the present invention takes is:
A kind of efficient super-resolution imaging device of district management, including: imaging lens group 1, for obtaining the optical signal of actual scene; High-resolution detector 2, is positioned on the back focal plane of imaging lens group 1, for gathering the optical signal that imaging lens group 1 is caught, obtains low-resolution image; Image storage module 3, the low-resolution image that storage high-resolution detector 2 obtains; Image super-resolution rebuilding module 5, for carrying out super-resolution rebuilding to the low-resolution image obtained; Image output display module 6, image after rebuilding for output and display;
Image pre-processing module 4 it is provided with between image storage module 3 and image super-resolution rebuilding module 5, this image pre-processing module 4 is made up of image denoising submodule 41, deblurring submodule 42 and sample piecemeal submodule 43, for the low-resolution image obtained on high-resolution detector 2 is carried out pretreatment, obtain low-resolution image to be reconstructed and dictionary training sample image; Image super-resolution rebuilding module 5 is managed submodule 51 and dictionary training by image-regionization and forms with compartmentalization reconstruction submodule 52, for the low-resolution image through pretreatment is carried out district management, sample training and image super-resolution rebuilding, obtain super-resolution image.
The efficient super-resolution imaging device of above-mentioned district management, imaging lens group 1 adopts varifocal imaging lens group.
The efficient super-resolution imaging device of above-mentioned district management, the low-resolution image after denoising and deblurring is carried out piecemeal by sample piecemeal submodule 43, obtains being sized to the sub-block of K × K, wherein 50≤K≤100, non-overlapping pixel between each sub-block.
The efficient super-resolution imaging device of above-mentioned district management, low-resolution image to be reconstructed is carried out region division by regional management module 51, obtains being sized to the subregion of W × W, wherein 100≤W≤150, pixel count overlapping between adjacent subarea territory is P, wherein 3≤P≤6.
The efficient super-resolution imaging device of above-mentioned district management, regional management module 51 adopts Canny operator edge detection method that subregion is carried out quantity of information judgement, obtains high information quantity region and Poor information region.
The efficient super-resolution imaging device of above-mentioned district management, dictionary training, with compartmentalization reconstruction submodule 52, high information quantity region is taken based on dictionary learning and sparse representation method carries out super-resolution rebuilding, and take bicubic interpolation method to carry out super-resolution rebuilding in Poor information region, obtain super-resolution rebuilding subregion.
The formation method of the efficient super-resolution imaging device of above-mentioned district management, comprises the steps:
Step 1: obtain the optical signal of actual scene;
Step 2: obtain low-resolution image;
Step 3: storage low-resolution image;
Step 4: low-resolution image is carried out pretreatment,
Step 4a): to low-resolution image denoising;
Step 4b): to step 4a) image deblurring that obtains, obtain low-resolution image to be reconstructed;
Step 4c): to low-resolution image denoising, deblurring and sample piecemeal, obtain dictionary training sample;
Step 5: low-resolution image to be reconstructed is carried out regional management,
Step 5a): low-resolution image to be reconstructed is carried out region division;
Step 5b): to step 5a) in the subregion that obtains carry out quantity of information judgement, obtain high information quantity region and Poor information region;
Step 6: dictionary training sample is carried out dictionary training, obtains high-resolution dictionary and low-resolution dictionary; High information quantity region and Poor information region are carried out super-resolution rebuilding, obtains super-resolution rebuilding subregion;
Step 7: super-resolution rebuilding subregion is carried out image mosaic, obtains super-resolution rebuilding image.
The formation method of the efficient super-resolution imaging device of above-mentioned district management, carries out dictionary training to dictionary training sample in step 6, comprises the steps:
Step 6a): dictionary training sample carrying out high-frequency characteristic extraction, adopts two-dimensional filtering operator filtering device group, this bank of filters is expressed as:
Wherein f1=[1-1], f2Being the two-dimentional LOG operator of 5 × 5, T representing matrix transposition operates;
Step 6b): the high-frequency characteristic extracted is carried out dictionary training, adopts off-line K-SVD algorithm, it is thus achieved that high-resolution dictionary DhWith low-resolution dictionary Dl, namely solve equation below:
Wherein xhFor high-frequency characteristic, α is rarefaction representation coefficient matrix, αiFor element each in matrix, T0For degree of rarefication, DhIt is required high-resolution dictionary, extracts high-frequency information to after down-sampled for sample image, adopt identical method can obtain low-resolution dictionary Dl。
The formation method of the efficient super-resolution imaging device of above-mentioned district management, the high information quantity regional reconstruction in step 6 adopts the method based on dictionary learning Yu rarefaction representation, comprises the steps:
Step 6c): adopt OMP Algorithm for Solving rarefaction representation coefficient, namely solve equation below:
Wherein, y is low resolution subregion to be reconstructed, and β is rarefaction representation coefficient;
Step 6d): by β and high-resolution dictionary DhIt is multiplied, obtains the super-resolution rebuilding subregion rebuild.
The formation method of the efficient super-resolution imaging device of above-mentioned district management, based on the method for dictionary learning Yu rarefaction representation, the low-resolution dictionary D usedlWith high-resolution dictionary Dh, it is possible to adopt existing training dictionary in off-line training result or device.
The present invention compared with prior art, has the advantage that
1, due to the image super-resolution rebuilding module of employing in the present invention, managed submodule and dictionary training by image-regionization to form with compartmentalization reconstruction submodule, achieve the super-resolution rebuilding of the district management of image to be reconstructed, the dictionary training of training sample and area image, effectively raise reconstructed image resolution, shorten imaging time.
2, due to the fact that and be provided with image pre-processing module between image storage module and image super-resolution rebuilding module, before image super-resolution rebuilding, the low-resolution image got on detector can be carried out denoising and deblurring processes, improve the picture quality of acquisition.
3, owing to dictionary training sample is a large amount of actual scene imaging results in the present invention, there is more high-frequency characteristic information, utilize low-resolution image self information as compared with the method for dictionary training sample with of the prior art, the quantity of information comprised in dictionary substantially increases, and further increases reconstructed image resolution.
4, the efficient super-resolution imaging device of the district management in the present invention adopts sample acquisition not changing in device under the premise of each locations of structures, dictionary training, the method of image super-resolution rebuilding integration obtains super-resolution image, with prior art adopts by regulating the detector position of camera in camera array with compared with obtaining the method for target area image, operation easier is low with processing cost, simultaneously again because device itself both can obtain scene image, training dictionary required in training sample and super-resolution rebuilding can be obtained again, need not be assisted by other devices, more convenient to operate.
5, owing to completing region division when low-resolution image to be reconstructed is carried out regional management in the present invention, all subregion independently carries out super-resolution rebuilding, achieve miniaturization image super-resolution rebuilding, significantly reduce the requirement to device internal memory in image reconstruction process.
Accompanying drawing explanation
Fig. 1 is the structural representation of the present invention;
Fig. 2 is the piecemeal schematic diagram of sample image in the present invention;
Fig. 3 is that in the present invention, pending low-resolution image region divides schematic diagram;
Fig. 4 is the flowage structure figure of super resolution ratio reconstruction method in the present invention;
Fig. 5 is contrast effect figure before and after super-resolution rebuilding in the present invention.
Detailed description of the invention
Below in conjunction with the drawings and specific embodiments, technical scheme is further described.
Embodiment 1
With reference to Fig. 1, the present invention includes imaging lens group 1, high-resolution detector 2, image storage module 3, image pre-processing module 4, image super-resolution rebuilding module 5, image show and output module 6, wherein high-resolution detector 2 is positioned on the back focal plane of imaging lens group 1, and all the other each modules are sequentially connected with; Image pre-processing module is made up of sample image denoising submodule 41, deblurring submodule 42 and sample piecemeal submodule 43, and image super-resolution rebuilding module 5 is managed module 51 and dictionary training by image-regionization and forms with compartmentalization reconstruction module 52.
Imaging lens group 1 adopts varifocal imaging lens group, because sample image needs scene information relative abundance, so adopting the longest focal length of lens group to obtain the optical signal of the abundant multiple outdoor scene scenes of detailed information; The optical signal collected is carried out imaging by detector 2, obtains low-resolution image; The low-resolution image obtained on detector 2 is stored by image storage module 3, in image storage module 3, the low-resolution image of storage can be divided into two classes: needs to carry out the target image of super-resolution rebuilding and carries out the source images required for sample training, because the reason of shooting condition and imaging system self, the low-resolution image of acquisition often exists serious image blurring and noise; The low resolution image that there is serious image blurring and noise is carried out corresponding pretreatment operation by image pre-processing module 4, wherein to needing the target image carrying out super-resolution rebuilding to carry out denoising through image denoising submodule 41, then in deblurring submodule 42, carry out deblurring process, obtain the low-resolution image to be reconstructed of correspondence, after the source images required for carrying out sample training is sequentially passed through the operation in image denoising submodule 41, image deblurring submodule 42 and sample piecemeal submodule 43, obtain dictionary training sample image;Low-resolution image to be reconstructed is carried out district management by the regional management module 51 in image super-resolution rebuilding module 5, employing has overlapped partitioning mode it is carried out image-region division and all subregion is carried out quantity of information judgement, obtain high information quantity region and Poor information region, then the dictionary training in image super-resolution rebuilding module 5 is rebuild in module 51 with compartmentalization and is utilized dictionary training sample image to carry out dictionary training, obtain high-resolution dictionary and corresponding low-resolution dictionary, and adopt diverse ways that high information quantity region and Poor information region are carried out super-resolution rebuilding, obtain super-resolution image.
With reference to Fig. 2, low-resolution image after image denoising and image deblurring is carried out piecemeal by the sample piecemeal submodule 43 in image pre-processing module 4, obtain being sized to the sub-block of K × K, wherein 50≤K≤100, the present embodiment K=80, region for undersize 80 × 80 is not subdivided, and final acquisition is smaller in size than in a large number or is equal to the sub-image of 80 × 80, as dictionary training sample image.
With reference to Fig. 3, low-resolution image to be reconstructed is carried out region division by the regional management module 51 in image super-resolution rebuilding module 5, wherein region dividing mode is for there being overlapping division, obtain being sized to the subregion of W × W, wherein 100≤W≤150, W=120 in the present embodiment, pixel count overlapping between adjacent subarea territory is P, wherein 3≤P≤6, P=5 in the present embodiment, final obtain several and be smaller in size than or subregion image equal to 120 × 120, the pixel count region less than 120 × 120 is no longer divided.
With reference to Fig. 4, the formation method of the efficient super-resolution imaging device of district management includes following step:
Step 1: obtained the optical signal of actual scene by imaging lens group 1;
Step 2: obtain low-resolution image on high-resolution detector 2;
Step 3: storage low-resolution image, it includes needing the target image carrying out super-resolution rebuilding and carrying out the source images required for sample training;
Step 4: low-resolution image is carried out pretreatment, wherein mainly includes image denoising, image deblurring and the operation of sample piecemeal, and its process step is:
Step 4a): to the low-resolution image denoising obtained in step 3;
Step 4b): to step 4a) image deblurring that obtains, obtain low-resolution image to be reconstructed;
Step 4c): to low-resolution image denoising, deblurring and sample piecemeal, obtain dictionary training sample;
Step 5: low-resolution image to be reconstructed is carried out regional management, divides, by region, the mode judged with quantity of information and obtains high information quantity region and Poor information region, and processing step is:
Step 5a): low-resolution image to be reconstructed is carried out region division;
Need the quantity of information of all subregion is judged after region divides, this patent adopts Canny operator edge detection method to carry out quantity of information judgement, first with Canny, all subregion is carried out rim detection, obtain the rim detection matrix that all subregion is corresponding, then element summation in matrix is carried out, detecting the subregion that quantity of information is maximum, be judged as Poor information region for the quantity of information subregion less than the 30% of maximum subregion quantity of information, all the other subregions are high information quantity region. Take bicubic interpolation algorithm to carry out N times of super-resolution rebuilding for Poor information region, and adopt the super resolution ratio reconstruction method based on dictionary learning and rarefaction representation to carry out N times of super-resolution rebuilding in high information quantity region.After super-resolution rebuilding, all subregion superposition image prime number becomes P × N.
Step 5b): to step 5a) in the subregion that obtains carry out quantity of information judgement, obtain high information quantity region and Poor information region;
Step 6: dictionary training sample is carried out dictionary training, obtains high-resolution dictionary and low-resolution dictionary; High information quantity region and Poor information region are carried out super-resolution rebuilding, obtains super-resolution rebuilding subregion;
Wherein dictionary training sample is carried out dictionary training, comprise the steps:
Step 6a): dictionary training sample carrying out high-frequency characteristic extraction, adopts two-dimensional filtering operator filtering device group, bank of filters used is f={f1,f2,f3,f4, it is made up of four different wave filter, is respectively as follows:
f1=[1 ,-1], f2=f1 T
f3=LOG, f3=f3 T
Wherein superscript T representing matrix transposition operation, LOG represents the two-dimensional filtering operator of a kind of 5 × 5. Extract through high-frequency characteristic and the sample high-frequency information that dictionary training is required after operation, can be obtained.
Step 6b): the high-frequency characteristic extracted is carried out dictionary training, adopts off-line K-SVD algorithm to carry out double; two sparse dictionary training, make xhRepresenting high-resolution sample high-frequency characteristic after feature extraction, α represents rarefaction representation coefficient matrix, αiFor element each in matrix, given initial dictionary D0, solve equation below:
By rarefaction representation factor alpha and dictionary DhIteration update (first time iterative process in Dh=D0), when the element in α is less than given numerical value degree of rarefication T0Time stop iteration, the high-resolution dictionary D now obtainedhIt is required high-resolution dictionary. Image in sample is carried out 1/3rd down-sampled, then carries out the image filtering operations that step is a kind of again, it is thus achieved that high-frequency information with xlRepresent, substitute in equation below:
Wherein DlRepresent low-resolution dictionary, given initial dictionary D0, pass through Dl(D in iteration for the first time is updated with the iteration of rarefaction representation factor alphal=D0), when the element in α is less than given numerical value degree of rarefication T0Time stop iteration, the low-resolution dictionary D now obtainedlIt is required low-resolution dictionary.
Wherein high information quantity regional reconstruction is adopted the method based on dictionary learning Yu rarefaction representation, comprise the steps:
Step 6c): adopt OMP Algorithm for Solving subregion y to be reconstructed at low-resolution dictionary DlUnder rarefaction representation factor beta, namely solve equation below:
Wherein T0For given degree of rarefication, βiFor the daughter element in matrix β.
Step 6d): rarefaction representation factor beta and the high-resolution dictionary D that will try to achievehIt is multiplied, obtains the super-resolution rebuilding subregion rebuild, it may be assumed that
X=Dhβ
Wherein X is the super-resolution subregion tried to achieve.
Step 7: super-resolution rebuilding subregion is carried out image mosaic, obtains super-resolution rebuilding image, because algorithm for reconstructing is N times of super-resolution rebuilding, so overlapping region pixel count is 5N, adopts the method that is averaging to merge overlapping partial pixel.
With reference to Fig. 5, in the present invention image super-resolution rebuilding method rebuild after the resolution of image apparently higher than the image resolution ratio without super-resolution rebuilding, owing to adopting subarea processing, therefore internal memory overflow problem in image reconstruction process is well solved, again because reasonably using bicubic interpolation algorithm, in addition training sample and the similarity of image information to be reconstructed, therefore rebuild efficiency and reconstructed image quality be all greatly promoted.
Embodiment 2
The structure of embodiment 2 is identical with embodiment 1, only following parameter is adjusted:
Sub-block size K=50, subregion size O=100, superposition image prime number P=3 between adjacent subarea territory.
Embodiment 3
The structure of embodiment 3 is identical with embodiment 1, only following parameter is adjusted:
Sub-block size K=100, subregion size O=150, superposition image prime number P=6 between adjacent subarea territory.
Above description and embodiment; it is only the preferred embodiment of the present invention; do not constitute any limitation of the invention; obviously for those skilled in the art; after having understood present invention and design principle; all be likely to when based on principles of the invention and structure, carry out in form and various corrections in details and change, but these based on the correction of inventive concept and change still within the scope of the claims of the present invention.
Claims (10)
1. an efficient super-resolution imaging device for district management, including: imaging lens group (1), for obtaining the optical signal of actual scene; High-resolution detector (2), is positioned on the back focal plane of imaging lens group (1), is used for gathering the optical signal that imaging lens group (1) is caught, and obtains low-resolution image; Image storage module (3), the upper low-resolution image obtained of storage high-resolution detector (2); Image super-resolution rebuilding module (5), for carrying out super-resolution rebuilding to the low-resolution image obtained; Image output display module (6), image after rebuilding for output and display;
It is characterized in that:
Image pre-processing module (4) it is provided with between described image storage module (3) and image super-resolution rebuilding module (5), this image pre-processing module (4) is made up of image denoising submodule (41), deblurring submodule (42) and sample piecemeal submodule (43), for the upper low-resolution image obtained of high-resolution detector (2) is carried out pretreatment, obtain low-resolution image to be reconstructed and dictionary training sample image; Described image super-resolution rebuilding module (5) is managed submodule (51) and dictionary training by image-regionization and forms with compartmentalization reconstruction submodule (52), for the low-resolution image through pretreatment is carried out district management, sample training and image super-resolution rebuilding, obtain super-resolution image.
2. the efficient super-resolution imaging device of district management according to claim 1, it is characterised in that described imaging lens group (1) adopts varifocal imaging lens group.
3. the efficient super-resolution imaging device of district management according to claim 1, it is characterized in that, low-resolution image after denoising and deblurring is carried out piecemeal by described sample piecemeal submodule (43), obtain being sized to the sub-block of K × K, wherein 50≤K≤100, non-overlapping pixel between each sub-block.
4. the efficient super-resolution imaging device of district management according to claim 1, it is characterized in that, low-resolution image to be reconstructed is carried out region division by described regional management module (51), obtain being sized to the subregion of W × W, wherein 100≤W≤150, pixel count overlapping between adjacent subarea territory is P, wherein 3≤P≤6.
5. the efficient super-resolution imaging device of the district management according to claim 1 or 4, it is characterized in that, described regional management module (51) adopts Canny operator edge detection method that subregion is carried out quantity of information judgement, obtains high information quantity region and Poor information region.
6. the efficient super-resolution imaging device of district management according to claim 1, it is characterized in that, described dictionary training, with compartmentalization reconstruction submodule (52), high information quantity region is taken based on dictionary learning and sparse representation method carries out super-resolution rebuilding, and take bicubic interpolation method to carry out super-resolution rebuilding in Poor information region, obtain super-resolution rebuilding subregion.
7. the formation method of the efficient super-resolution imaging device of district management according to claim 1, comprises the steps:
1) optical signal of actual scene is obtained;
2) low-resolution image is obtained;
3) storage low-resolution image;
4) low-resolution image is carried out pretreatment,
4a) to low-resolution image denoising;
4b) to 4a) image deblurring that obtains, obtain low-resolution image to be reconstructed;
4c) to low-resolution image denoising, deblurring and sample piecemeal, obtain dictionary training sample;
5) low-resolution image to be reconstructed is carried out regional management,
5a) low-resolution image to be reconstructed is carried out region division;
5b) to 5a) in the subregion that obtains carry out quantity of information judgement, obtain high information quantity region and Poor information region;
6) dictionary training sample is carried out dictionary training, obtain high-resolution dictionary and low-resolution dictionary; High information quantity region and Poor information region are carried out super-resolution rebuilding, obtains super-resolution rebuilding subregion;
7) super-resolution rebuilding subregion is carried out image mosaic, obtain super-resolution rebuilding image.
8. the formation method of the efficient super-resolution imaging device of district management according to claim 7, it is characterised in that described 6) in dictionary training sample is carried out dictionary training, comprise the steps:
6a) dictionary training sample carrying out high-frequency characteristic extraction, adopt two-dimensional filtering operator filtering device group, this bank of filters is expressed as
Wherein f1=[1-1], f2Being the two-dimentional LOG operator of 5 × 5, T representing matrix transposition operates;
6b) high-frequency characteristic extracted is carried out dictionary training, adopt K-SVD algorithm, it is thus achieved that high-resolution dictionary DhWith low-resolution dictionary Dl, namely solve equation below:
Wherein xhFor high-frequency characteristic, α is rarefaction representation coefficient matrix, αiFor element each in matrix, T0For degree of rarefication, DhIt is required high-resolution dictionary, extracts high-frequency information to after down-sampled for sample image, adopt identical method can obtain low-resolution dictionary Dl。
9. the formation method of the efficient super-resolution imaging device of district management according to claim 7, it is characterised in that described 6) in high information quantity regional reconstruction adopt based on the method for dictionary learning Yu rarefaction representation, comprise the steps:
6c) adopt OMP Algorithm for Solving rarefaction representation coefficient, namely solve equation below:
Wherein, y is low resolution subregion to be reconstructed, and β is rarefaction representation coefficient, βiFor the daughter element in matrix β;
6d) by β and high-resolution dictionary DhIt is multiplied, obtains the super-resolution rebuilding subregion rebuild.
10. the formation method of the efficient super-resolution imaging device of district management according to claim 9, it is characterised in that the described method based on dictionary learning Yu rarefaction representation, the low-resolution dictionary D usedlWith high-resolution dictionary Dh, it is possible to adopt existing training dictionary in off-line training result or device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610033903.9A CN105678728A (en) | 2016-01-19 | 2016-01-19 | High-efficiency super-resolution imaging device and method with regional management |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610033903.9A CN105678728A (en) | 2016-01-19 | 2016-01-19 | High-efficiency super-resolution imaging device and method with regional management |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105678728A true CN105678728A (en) | 2016-06-15 |
Family
ID=56301486
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610033903.9A Pending CN105678728A (en) | 2016-01-19 | 2016-01-19 | High-efficiency super-resolution imaging device and method with regional management |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105678728A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106204621A (en) * | 2016-07-22 | 2016-12-07 | 北京航空航天大学 | The accurate extracting method of high-definition picture object that a kind of self adaptation is divided and ruled |
| CN106960416A (en) * | 2017-03-20 | 2017-07-18 | 武汉大学 | A kind of video satellite compression image super-resolution method of content complexity self adaptation |
| CN107093170A (en) * | 2017-04-21 | 2017-08-25 | 西安电子科技大学 | A kind of real-time super-resolution rate method for reconstructing |
| CN107133916A (en) * | 2017-04-21 | 2017-09-05 | 西安科技大学 | Image-scaling method |
| CN107133915A (en) * | 2017-04-21 | 2017-09-05 | 西安科技大学 | A kind of image super-resolution reconstructing method based on study |
| CN108665509A (en) * | 2018-05-10 | 2018-10-16 | 广东工业大学 | A kind of ultra-resolution ratio reconstructing method, device, equipment and readable storage medium storing program for executing |
| CN109257523A (en) * | 2018-10-19 | 2019-01-22 | 江西财经大学 | A kind of medical image processing devices and image processing method |
| CN109785235A (en) * | 2018-12-29 | 2019-05-21 | 华中光电技术研究所（中国船舶重工集团有限公司第七一七研究所） | Compressed sensing based infrared pedestrian image super resolution ratio reconstruction method and system |
| CN110288530A (en) * | 2019-06-28 | 2019-09-27 | 北京金山云网络技术有限公司 | A kind of pair of image carries out the processing method and processing device of super-resolution rebuilding |
| CN110298790A (en) * | 2019-06-28 | 2019-10-01 | 北京金山云网络技术有限公司 | A kind of pair of image carries out the processing method and processing device of super-resolution rebuilding |
| CN111145202A (en) * | 2019-12-31 | 2020-05-12 | 北京奇艺世纪科技有限公司 | Model generation method, image processing method, device, equipment and storage medium |
| CN111626931A (en) * | 2020-05-06 | 2020-09-04 | Oppo广东移动通信有限公司 | Image processing method, image processing apparatus, storage medium, and electronic device |
| CN112131857A (en) * | 2020-09-11 | 2020-12-25 | 安徽中科新辰技术有限公司 | Ultrahigh-resolution visual self-adaptive typesetting method |
| CN113727050A (en) * | 2021-11-04 | 2021-11-30 | 山东德普检测技术有限公司 | Video super-resolution processing method and device for mobile equipment and storage medium |
| WO2023274404A1 (en) * | 2021-07-01 | 2023-01-05 | Beijing Bytedance Network Technology Co., Ltd. | Application of super resolution |
| CN112131857B (en) * | 2020-09-11 | 2024-05-31 | 安徽中科新辰技术有限公司 | Self-adaptive typesetting method for ultrahigh resolution visualization |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930518A (en) * | 2012-06-13 | 2013-02-13 | 上海汇纳网络信息科技有限公司 | Improved sparse representation based image super-resolution method |
| CN103077505A (en) * | 2013-01-25 | 2013-05-01 | 西安电子科技大学 | Image super-resolution reconstruction method based on dictionary learning and structure clustering |
| US20130170767A1 (en) * | 2012-01-04 | 2013-07-04 | Anustup Kumar CHOUDHURY | Image content enhancement using a dictionary technique |
| CN103455988A (en) * | 2013-09-17 | 2013-12-18 | 西安电子科技大学 | Super-resolution image reconstruction method based on structure self-similarity and sparse representation |
| CN104899830A (en) * | 2015-05-29 | 2015-09-09 | 清华大学深圳研究生院 | Image super-resolution method |
-
2016
- 2016-01-19 CN CN201610033903.9A patent/CN105678728A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130170767A1 (en) * | 2012-01-04 | 2013-07-04 | Anustup Kumar CHOUDHURY | Image content enhancement using a dictionary technique |
| CN102930518A (en) * | 2012-06-13 | 2013-02-13 | 上海汇纳网络信息科技有限公司 | Improved sparse representation based image super-resolution method |
| CN103077505A (en) * | 2013-01-25 | 2013-05-01 | 西安电子科技大学 | Image super-resolution reconstruction method based on dictionary learning and structure clustering |
| CN103455988A (en) * | 2013-09-17 | 2013-12-18 | 西安电子科技大学 | Super-resolution image reconstruction method based on structure self-similarity and sparse representation |
| CN104899830A (en) * | 2015-05-29 | 2015-09-09 | 清华大学深圳研究生院 | Image super-resolution method |
Non-Patent Citations (3)
| Title |
|---|
| YANG JIAN CHAO 等: "Image super-resolution via sparse representation", 《IEEE TRANSACTION ON IMAGE PROCESSING》 * |
| 侯鲜桃: "超分辨率图像重建算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 熊卫辉: "基于稀疏表示和字典学习的单帧图像超分辨率算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106204621B (en) * | 2016-07-22 | 2019-03-08 | 北京航空航天大学 | A kind of accurate extracting method of high-definition picture object adaptively divided and ruled |
| CN106204621A (en) * | 2016-07-22 | 2016-12-07 | 北京航空航天大学 | The accurate extracting method of high-definition picture object that a kind of self adaptation is divided and ruled |
| CN106960416A (en) * | 2017-03-20 | 2017-07-18 | 武汉大学 | A kind of video satellite compression image super-resolution method of content complexity self adaptation |
| CN106960416B (en) * | 2017-03-20 | 2019-05-10 | 武汉大学 | A kind of video satellite that content complexity is adaptive compression image super-resolution method |
| CN107093170B (en) * | 2017-04-21 | 2020-05-12 | 西安电子科技大学 | Real-time super-resolution reconstruction method |
| CN107093170A (en) * | 2017-04-21 | 2017-08-25 | 西安电子科技大学 | A kind of real-time super-resolution rate method for reconstructing |
| CN107133916A (en) * | 2017-04-21 | 2017-09-05 | 西安科技大学 | Image-scaling method |
| CN107133915A (en) * | 2017-04-21 | 2017-09-05 | 西安科技大学 | A kind of image super-resolution reconstructing method based on study |
| CN108665509A (en) * | 2018-05-10 | 2018-10-16 | 广东工业大学 | A kind of ultra-resolution ratio reconstructing method, device, equipment and readable storage medium storing program for executing |
| CN109257523A (en) * | 2018-10-19 | 2019-01-22 | 江西财经大学 | A kind of medical image processing devices and image processing method |
| CN109785235A (en) * | 2018-12-29 | 2019-05-21 | 华中光电技术研究所（中国船舶重工集团有限公司第七一七研究所） | Compressed sensing based infrared pedestrian image super resolution ratio reconstruction method and system |
| CN110288530A (en) * | 2019-06-28 | 2019-09-27 | 北京金山云网络技术有限公司 | A kind of pair of image carries out the processing method and processing device of super-resolution rebuilding |
| CN110298790A (en) * | 2019-06-28 | 2019-10-01 | 北京金山云网络技术有限公司 | A kind of pair of image carries out the processing method and processing device of super-resolution rebuilding |
| CN111145202A (en) * | 2019-12-31 | 2020-05-12 | 北京奇艺世纪科技有限公司 | Model generation method, image processing method, device, equipment and storage medium |
| CN111145202B (en) * | 2019-12-31 | 2024-03-08 | 北京奇艺世纪科技有限公司 | Model generation method, image processing method, device, equipment and storage medium |
| CN111626931A (en) * | 2020-05-06 | 2020-09-04 | Oppo广东移动通信有限公司 | Image processing method, image processing apparatus, storage medium, and electronic device |
| CN111626931B (en) * | 2020-05-06 | 2023-08-08 | Oppo广东移动通信有限公司 | Image processing method, image processing device, storage medium and electronic apparatus |
| CN112131857A (en) * | 2020-09-11 | 2020-12-25 | 安徽中科新辰技术有限公司 | Ultrahigh-resolution visual self-adaptive typesetting method |
| CN112131857B (en) * | 2020-09-11 | 2024-05-31 | 安徽中科新辰技术有限公司 | Self-adaptive typesetting method for ultrahigh resolution visualization |
| WO2023274404A1 (en) * | 2021-07-01 | 2023-01-05 | Beijing Bytedance Network Technology Co., Ltd. | Application of super resolution |
| CN113727050A (en) * | 2021-11-04 | 2021-11-30 | 山东德普检测技术有限公司 | Video super-resolution processing method and device for mobile equipment and storage medium |
| CN113727050B (en) * | 2021-11-04 | 2022-03-01 | 山东德普检测技术有限公司 | Video super-resolution processing method and device for mobile equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105678728A (en) | High-efficiency super-resolution imaging device and method with regional management | |
| Chen et al. | Real-world single image super-resolution: A brief review | |
| Yan et al. | Attention-guided network for ghost-free high dynamic range imaging | |
| CN111080724B (en) | Fusion method of infrared light and visible light | |
| Jo et al. | Deep video super-resolution network using dynamic upsampling filters without explicit motion compensation | |
| Liu et al. | Video super-resolution based on deep learning: a comprehensive survey | |
| Jiang et al. | Learning spatial-spectral prior for super-resolution of hyperspectral imagery | |
| Zheng et al. | Crossnet: An end-to-end reference-based super resolution network using cross-scale warping | |
| CN107240066A (en) | Image super-resolution rebuilding algorithm based on shallow-layer and deep layer convolutional neural networks | |
| CN102982520B (en) | Robustness face super-resolution processing method based on contour inspection | |
| CN109447930B (en) | Wavelet domain light field full-focusing image generation algorithm | |
| DE112011102647T5 (en) | Method for producing super-resolution images and non-linear filter for its implementation | |
| CN112184604B (en) | Color image enhancement method based on image fusion | |
| CN102609931B (en) | Field depth expanding method and device of microscopic image | |
| Sun et al. | Deep maximum a posterior estimator for video denoising | |
| CN111986106A (en) | High dynamic image reconstruction method based on neural network | |
| Chen et al. | Image denoising via deep network based on edge enhancement | |
| Liu et al. | Learning noise-decoupled affine models for extreme low-light image enhancement | |
| Yanshan et al. | OGSRN: Optical-guided super-resolution network for SAR image | |
| Gong et al. | Learning deep resonant prior for hyperspectral image super-resolution | |
| Mehta et al. | Gated multi-resolution transfer network for burst restoration and enhancement | |
| CN111105354A (en) | Depth image super-resolution method and device based on multi-source depth residual error network | |
| CN113269672B (en) | Super-resolution cell image construction method and system | |
| CN110378850A (en) | A kind of zoom image generation method of combination Block- matching and neural network | |
| Xu et al. | Deformable kernel convolutional network for video extreme super-resolution |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160615 |
|
| WD01 | Invention patent application deemed withdrawn after publication |