CN103632357B - A kind of image super-resolution Enhancement Method separated based on illumination - Google Patents
A kind of image super-resolution Enhancement Method separated based on illumination Download PDFInfo
- Publication number
- CN103632357B CN103632357B CN201310037260.1A CN201310037260A CN103632357B CN 103632357 B CN103632357 B CN 103632357B CN 201310037260 A CN201310037260 A CN 201310037260A CN 103632357 B CN103632357 B CN 103632357B
- Authority
- CN
- China
- Prior art keywords
- image
- low
- resolution
- illumination
- frequency information
- 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
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image super-resolution Enhancement Method separated based on illumination, the method first carries out illumination separating treatment to reference picture, then the corresponding relation between a large amount of reference picture study image low-frequency information and high-frequency information is utilized, finally utilize this corresponding relation to obtain the high-frequency information of image to be reinforced, obtain high-definition picture.The present invention has taken into full account the ambient change impact on image procossing, greatly improves the effect that super-resolution image strengthens；Image for same scene, it is only necessary to the image under a light, significantly reduces the scale in learning sample storehouse；Simultaneously because the reduction of the scale of Sample Storehouse, the scale in the priori storehouse that training obtains also decreases, therefore when carrying out high dimensional data retrieval, can be time-consuming, and improve operation efficiency.
Description
Technical field
The invention belongs to technical field of image information processing, specifically, relate to a kind of image super-resolution Enhancement Method separated based on illumination, a kind of super-resolution processing method being recovered full resolution pricture by low-resolution image separated based on illumination.
Background technology
Along with the Internet and the fast development of the information processing technology, people are more and more higher to the requirement of image resolution ratio.But owing to being limited by physics image-forming condition and collection environment, the image resolution ratio collected is the lowest, and the visual effect of image is difficult to meet the needs of people.
Super-resolution image Enhancement Method utilizes character and the related information of adjacent image block of image itself, by certain algorithm process, from low resolution low quality input picture, high-resolution high quality graphic is estimated not increasing in the case of hardware puts into, image resolution ratio is made to exceed the resolution of optical hardware equipment, obtain the detailed information that image is potential, strengthen the availability of image.This method, once proposition, just causes extensively paying attention to and paying close attention to of many scholars.Harris [refer to document J.LHarris.DiffractionandResolvingPower.JournalofOpticalS ocietyofAmerican, Vol.54, No.7, pp.931-936,1964] prove theoretically, two different objects will not produce same image, and therefore, any image obtained under noise free conditions is the most corresponding with an object.So, according to the low-resolution image of an object, the details of this object being strengthened existence may.Through the follow-up study of scientific research personnel, super-resolution image Enhancement Method has been achieved with more result, and its feasibility is the most undoubted.
Super-resolution image Enhancement Method has a wide range of applications in multiple fields such as vision monitoring, public safety, remote sensing, medical imaging and high-definition televisions.In terms of satellite remote sensing, owing to the resolution of image-forming condition with imaging device all has certain limitations, in general it is difficult to obtain the image of fine definition, therefore, is of great significance to generate high-definition picture tool from the low-resolution image obtained；In Video transmission system, by expanded view picture and increase details, super-resolution still image can be obtained from video sequence, common video signal is changed into high-definition TV signal；At public safety field, process the low-quality image caused due to move, defocus, environment etc., obtain high quality graphic；In living things feature recognition field, can obtain high-resolution facial image by Super-resolution Image Restoration method, this high-resolution human face image is used for recognition of face, is greatly improved accuracy of identification.
Super-resolution Enhancement Method has become as one of research topic that image processing field is the most active in the world.Super-resolution Enhancement Method substantially can be divided into both direction: based on the method reconstructed and method based on study.The main thought of super-resolution Enhancement Method based on reconstruct is to utilize multiframe to repeat to take pictures the redundancy of image, reconstructs super-resolution image, eliminates and reduce mixing effect.Super-resolution Enhancement Method based on reconstruct needs to set up the imaging model tallied with the actual situation and the model that degrades and it needs to image sequence is carried out accurate sub-pel motion estimation, and this is the most extremely difficult in actual treatment.Super-resolution Enhancement Method based on study is to grow up in the case of method based on reconstruct is met difficulty, although starting late, but appear at present make up a lot of not enough of method based on reconstruct, the development of super-resolution Enhancement Method combined with intelligent technology based on study, for certain types of image, effectively utilize priori, spatial resolution can be effectively improved, it it is the direction being worth research further, the present invention is just belonging to super-resolution field based on study, to solve the key issue in this field.
Super-resolution Enhancement Method based on study [be refer to document S.Baker by the Baker etc. of Ka Naiji-Mei Long laboratory the earliest, andT.Kanade.LimitsonSuper-resolutionandHowtoBreakThem.Pr oc.ofComputerVisionandPatternRecognition, pp.372-379,2000] propose, they propose a kind of method carrying out strengthening based on priori, utilize machine learning algorithm to go training to specify classification image, the priori obtained is used for super-resolution and strengthens.Massachusetts Institute of Technology Freeman etc. [refer to document W.T.Freeman, T.R.Jones, andE.C.Pasztor.Example-BasedSuper-Resolution.IEEECompute rGraphicsandApplications, Vol.22, No.2, PP.56-65,2002] method based on sample is proposed, utilize markov network to carry out in learning training storehouse the relation between low-resolution image and high-definition picture block, with study to relation predict the detailed information of input low-resolution image.Bishop etc. [refer to document C.M.Bishop, A.Blake, andB.Marthi.Super-ResolutionEnhancementofVideo.Proc.ofth e9thInternationalConferenceonArtificialIntelligenceandSt atistics, pp.410-414,2003] utilize an image block data storehouse to the relation obtaining between the Mid Frequency of natural image and high band, and the training set strengthened as present image by a part for the super-resolution image being enhanced, thus with the addition of additional high-frequency information.Joshi etc. [refer to document M.V.Joshi, S.Chaudhuri, andR.Panuganti.ALearning-BasedMethodforImageSuper-Resolu tionfromZoomedObservations.IEEETransactionsonSystems, ManandCybernetics, Vol.35, No.3, pp.527-537,2005] propose a kind of to merge the method that various sizes of image carries out resolution enhancing.Super-resolution Enhancement Method based on study needs to search for the information of coupling from extensive prior data bank, and the calculating resource of needs is very big, lacks the research of this respect, causes elapsed time long, have impact on the application of the method in conventional research work.
The basic ideas of super-resolution Enhancement Method based on study are, utilize the corresponding relation between a large amount of reference picture study image low-frequency information and high-frequency information, utilize this corresponding relation to obtain the high-frequency information of image to be reinforced, obtain high-definition picture.The method is the most in two steps, off-line training process and strengthen process online, wherein, off-line training process is the method utilizing statistical learning, analyze the image reference storehouse being made up of a large amount of high-definition pictures and low-resolution image, setting up the corresponding relation between the low-frequency information in low-resolution image and high-frequency information, build knowledge base, its processing procedure is as shown in Figure 1；Online strengthen during, to input picture piecemeal, extract search characteristics vector for each image block, knowledge base searched for the high-frequency information mated most, after merging, obtains high-definition picture, its processing procedure is as shown in Figure 2.
In above-mentioned existing image enchancing method, do not account for the impact of image light, but in actual applications, owing to light change causes the image change of Same Scene obtained various, the Image Priori Knowledge storehouse developed under a certain light condition and character representation method, can not apply well under another light condition, this just greatly limit the range of application of super-resolution image Enhancement Method based on study.Therefore, in the urgent need to a kind of image enchancing method not affected by light change, utilize limited training sample, while strengthening image detail, keep the continuity of the illumination of image.The present invention, on the basis of existing processing method, adds the processing links that illumination separates, it is possible to the above-mentioned problems of the prior art of effective solution.
Summary of the invention
In order to solve above-mentioned problems of the prior art, the present invention proposes a kind of image super-resolution Enhancement Method separated based on illumination.The method comprises the following steps:
Step 1, sets up the training image storehouse containing several high-definition pictures；
Step 2, carries out illumination separating treatment to each panel height image in different resolution in described training image storehouse, obtains illumination invariant image and light image；
Step 3, the described illumination invariant image obtaining described step 2 degrades and processes the low-resolution image obtaining correspondence；
Step 4, the corresponding low-resolution image obtained described high-definition picture with described step 3 is asked poor, is obtained high-frequency information image；
Step 5, the high-frequency information image that the low-resolution image obtained for described step 3 and described step 4 obtain, described low-resolution image and described high-frequency information image are divided into high-frequency information image block and the low-resolution image block of multiple overlap, correspondence position according to image block, set up multiple high-low-resolution image block pair, and respectively high-frequency information image block and low-resolution image block being extracted high-frequency characteristic vector sum characteristics of low-frequency vector, high-frequency characteristic vector sum characteristics of low-frequency extraction obtained vector is to composition priori storehouse；
Step 6, expands low-resolution image to be reinforced to enhanced size；
Step 7, utilizes illumination method for separating and processing, the low-resolution image after expanding is carried out illumination pretreatment, obtains illumination invariant image and light image；
Step 8, is divided into the illumination invariant image obtained in described step 7 image block of multiple overlap, and each image block extracts the characteristic vector identical with the feature extracted in described step 5；
Step 9, extract, based on described step 8, the characteristic vector that obtains inquire about in described priori storehouse and obtain extracting with described step 8 multiple characteristic vectors that the characteristic vector that obtains is mated most, thus obtain the high-frequency information image block of correspondence according to the corresponding relation between described characteristic vector or described characteristic vector；
The illumination invariant image overlay obtained in step 10, the high-frequency information image block described step 9 obtained and described step 7 merges, and obtains the high-definition picture of illumination invariant；
Step 11, carries out image co-registration to the high-definition picture of the light image obtained in described step 7 Yu described illumination invariant, obtains enhanced high-definition picture.
The inventive method application is extensive, particularly in fields such as public security, monitoring, remote sensing, owing to prior art is limited by acquisition condition, there is the problems such as resolution is the highest, fuzzy sometimes in the image collected, therefore significantly limit the actually used value of image, and the inventive method can promote the resolution of image effectively, strengthen the data value of image.
Accompanying drawing explanation
Fig. 1 is the off-line training process flow diagram flow chart of super-resolution Enhancement Method in prior art；
Fig. 2 be in prior art super-resolution Enhancement Method strengthen process flow diagram flow chart online；
Fig. 3 shows the off-line training process of image super-resolution Enhancement Method according to an embodiment of the invention；
Fig. 4 shows the online enhancing process of image super-resolution Enhancement Method according to an embodiment of the invention.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and referring to the drawings, the present invention is described in more detail.
The definition of super-resolution is to be rebuild high-definition picture by low-resolution image.Given single width low-resolution image, is divided into the image block of overlap, utilizes priori storehouse, estimate the high-definition picture block of each image block according to setting scheme, is then passed through merging the high-definition picture obtaining strengthening.Lacking in low-resolution image and describe details high-frequency information, therefore, the purpose of enhancing namely utilizes priori storehouse, obtains high-frequency information, is superposed with low-resolution image by high-frequency information thus produce the image of enhancing.
The present invention is directed to method of the prior art and affected problem bigger than normal by light, add the link that illumination separates in processing procedure, utilizing the information of illumination invariant in image is foundation, obtains high-frequency information, is finally further added by Lighting information, reduction real scene.The inventive method is divided into two processes: off-line training process and strengthen process online.The general off-line of previous process is carried out, and builds effective priori storehouse；Later process is carried out online, for the image of input, quickly obtains the image strengthened.The present invention, during off-line training process and online enhancing, all adds the processing procedure that illumination separates, to reduce the impact that super-resolution image is strengthened by illumination.
Fig. 3 shows the off-line training process of image super-resolution Enhancement Method according to an embodiment of the invention.The purpose of described off-line training process is to build a priori storehouse, and in the present invention, described priori storehouse is obtained by one group of high-resolution image set training.
As it is shown on figure 3, the off-line training process of image super-resolution Enhancement Method of the present invention includes following step:
Step 1, sets up the training image storehouse containing several high-definition pictures；
Step 2, carries out illumination separating treatment to each panel height image in different resolution in described training image storehouse, obtains illumination invariant image and light image；
In view of illumination variation, the impact of image is typically the overall situation, which determines the basic keynote of image chroma and brightness.nullLand [refer to document E.H.Land.AnAlternativeTechniquefortheComputationoftheDes ignatorintheRetinexTheoryofColorVision.Proc.ofNationalAc ademyofScience:NationalAcademyofScience，UAS，83，Pp:3078-3080，1986] a model Retinex algorithm how regulating color and the brightness perceiving object about human visual system is proposed，Its a kind of image processing algorithm based on illumination compensation，Can be in gray scale dynamic range compression，Edge strengthens and color constancy three aspect reaches balance.By this algorithm, can be with the light image of separating reaction ambient and the illumination invariant image of reaction object reflective nature.The present invention utilizes this algorithm that high-definition picture is carried out illumination separating treatment exactly, obtains illumination invariant image and light image.
Step 3, the described illumination invariant image obtaining described step 2 degrades and processes the low-resolution image obtaining correspondence；
In this step, described in the process that degrades can use multiple method of the prior art, such as can first use Gaussian smoothing, then carry out down-sampled, but finally utilize interpolation method to obtain the image that same size resolution is low.
Step 4, the corresponding low-resolution image obtained described high-definition picture with described step 3 is asked poor, is obtained high-frequency information image；
Step 5, the high-frequency information image that the low-resolution image obtained for described step 3 and described step 4 obtain, use identical image block strategy, described low-resolution image and described high-frequency information image are divided into high-frequency information image block and the low-resolution image block of multiple overlap, correspondence position according to image block, set up multiple high-low-resolution image block pair, and respectively high-frequency information image block and low-resolution image block are extracted high-frequency characteristic vector sum characteristics of low-frequency vector, high-frequency characteristic vector sum characteristics of low-frequency vector extraction obtained is to composition priori storehouse；
In this step, described high-frequency characteristic vector is typically directly made up of image information corresponding in high-frequency information image block, and described characteristics of low-frequency vector can be made up of the gray feature of corresponding low-resolution image block, Gradient Features, can also be made up of the statistical nature of image block, such as rectangular histogram etc., described high-frequency characteristic vector sum characteristics of low-frequency vector constitutes the locating vector of respective image block.
The purpose of super-resolution Enhancement Method of the present invention is to recover that high-definition picture possesses and that low-resolution image lacks high-frequency information.Therefore, the most valuable in training image is the difference information i.e. high-frequency information image of low-resolution image and original high-resolution image, and the corresponding relation between difference information and low-resolution image.Above-mentioned off-line training process is exactly based on priori storehouse and establishes this relation.
Fig. 4 shows the online enhancing process of image super-resolution Enhancement Method according to an embodiment of the invention.As shown in Figure 4, described online enhancing process includes following step:
Step 6, by low-resolution image to be reinforced, such as by interpolation method, expands enhanced size to；
Step 7, utilizes illumination method for separating and processing, the low-resolution image after expanding is carried out illumination pretreatment, obtains illumination invariant image and light image；
Step 8, for the illumination invariant image obtained in described step 7, according to being divided into the image block of multiple overlap with identical method during training, and extracts the characteristic vector identical with the feature extracted in described step 5 to each image block；
Step 9, extract, based on described step 8, the characteristic vector that obtains inquire about in described priori storehouse and obtain extracting with described step 8 multiple characteristic vectors that the characteristic vector that obtains is mated most, thus obtain the high-frequency information image block of correspondence according to the corresponding relation between described characteristic vector or described characteristic vector；
Characteristic vector matching process in this step, can use traditional vector similarity computational methods, such as carry out vector similarity calculating according to Euclidean distance, weighted euclidean distance, mahalanobis distance etc..
The illumination invariant image overlay obtained in step 10, the high-frequency information image block described step 9 obtained and described step 7 merges, and obtains the high-definition picture of illumination invariant；
Image overlay fusion method in this step can use the fusion method of Pixel-level, the image interfusion method such as average the most of the prior art, weighted sum.
Step 11, carries out image co-registration to the high-definition picture of the light image obtained in described step 7 Yu described illumination invariant, obtains enhanced high-definition picture.
Advantage acquired by the inventive method has following aspects:
1) effect that super-resolution image strengthens is promoted
The present invention can take into full account the ambient change impact on image procossing, thus greatly improves the effect that super-resolution image strengthens.
2) Sample Storehouse of study is reduced
The present invention is for the image of same scene, it is only necessary to the image under a light, thus significantly reduces the scale in learning sample storehouse.
3) computational efficiency is improved
Due to the fact that the reduction of the scale of Sample Storehouse, the scale in the priori storehouse that training obtains also decrease, therefore when carrying out high dimensional data retrieval, can be time-consuming, improve operation efficiency.
Particular embodiments described above; the purpose of the present invention, technical scheme and beneficial effect are further described; it is it should be understood that; the foregoing is only the specific embodiment of the present invention; it is not limited to the present invention; all within the spirit and principles in the present invention, any modification, equivalent substitution and improvement etc. done, should be included within the scope of the present invention.
Claims (10)
1. the image super-resolution Enhancement Method separated based on illumination, it is characterised in that the method comprises the following steps:
Step 1, sets up the training image storehouse containing several high-definition pictures；
Step 2, carries out illumination separating treatment to each panel height image in different resolution in described training image storehouse, obtains illumination invariant image and light image；
Step 3, the described illumination invariant image obtaining described step 2 degrades and processes the low-resolution image obtaining correspondence；
Step 4, the corresponding low-resolution image obtained described illumination invariant high-definition picture after illumination separates with described step 3 is asked poor, is obtained high-frequency information image；
Step 5, the high-frequency information image that the low-resolution image obtained for described step 3 and described step 4 obtain, described low-resolution image and described high-frequency information image are divided into high-frequency information image block and the low-resolution image block of multiple overlap, correspondence position according to image block, set up multiple high-low-resolution image block pair, and respectively high-frequency information image block and low-resolution image block being extracted high-frequency characteristic vector sum characteristics of low-frequency vector, high-frequency characteristic vector sum characteristics of low-frequency extraction obtained vector is to composition priori storehouse；
Step 6, expands low-resolution image to be reinforced to enhanced size；
Step 7, utilizes illumination method for separating and processing, the low-resolution image after expanding is carried out illumination pretreatment, obtains illumination invariant image and light image；
Step 8, is divided into the illumination invariant image obtained in described step 7 image block of multiple overlap, and each image block extracts the characteristic vector identical with the feature extracted in described step 5；
Step 9, extract, based on described step 8, the characteristic vector that obtains inquire about in described priori storehouse and obtain extracting with described step 8 multiple characteristic vectors that the characteristic vector that obtains is mated most, thus obtain the high-frequency information image block of correspondence according to the corresponding relation between described characteristic vector or described characteristic vector；
The illumination invariant image overlay obtained in step 10, the high-frequency information image block described step 9 obtained and described step 7 merges, and obtains the high-definition picture of illumination invariant；
Step 11, carries out image co-registration to the high-definition picture of the light image obtained in described step 7 Yu described illumination invariant, obtains enhanced high-definition picture.
Method the most according to claim 1, it is characterised in that in described step 3, uses Gaussian smoothing, down-sampled and interpolation to obtain described low-resolution image successively.
Method the most according to claim 1, it is characterised in that described low-resolution image and the described same size of illumination invariant image.
Method the most according to claim 1, it is characterised in that in step 5, described high-frequency characteristic vector is directly made up of image information corresponding in high-frequency information image block.
Method the most according to claim 1, it is characterised in that in step 5, described characteristics of low-frequency vector is made up of gray feature, Gradient Features or the statistical nature of corresponding low-resolution image block.
Method the most according to claim 5, it is characterised in that described statistical nature is rectangular histogram.
Method the most according to claim 1, it is characterised in that expand low-resolution image to be reinforced by interpolation method in described step 6.
Method the most according to claim 1, it is characterised in that in described step 9, uses vector similarity computational methods to mate characteristic vector.
Method the most according to claim 8, it is characterised in that described vector similarity computational methods are for carry out vector similarity calculating according to Euclidean distance, weighted euclidean distance or mahalanobis distance.
Method the most according to claim 1, it is characterised in that in described step 10 and described step 11, uses the fusion method of Pixel-level to merge image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310037260.1A CN103632357B (en) | 2013-01-30 | 2013-01-30 | A kind of image super-resolution Enhancement Method separated based on illumination |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310037260.1A CN103632357B (en) | 2013-01-30 | 2013-01-30 | A kind of image super-resolution Enhancement Method separated based on illumination |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103632357A CN103632357A (en) | 2014-03-12 |
| CN103632357B true CN103632357B (en) | 2016-08-03 |
Family
ID=50213374
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310037260.1A Active CN103632357B (en) | 2013-01-30 | 2013-01-30 | A kind of image super-resolution Enhancement Method separated based on illumination |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103632357B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780318B (en) * | 2015-11-25 | 2020-07-14 | 北京大学 | Face magic structure method and face magic structure system |
| CN108460722A (en) * | 2018-01-31 | 2018-08-28 | 中国科学院上海技术物理研究所 | A kind of high-resolution wide visual field rate remotely sensed image method and device |
| CN111369475B (en) * | 2020-03-26 | 2023-06-23 | 北京百度网讯科技有限公司 | Method and apparatus for processing video |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101923721A (en) * | 2010-08-31 | 2010-12-22 | 汉王科技股份有限公司 | Non-illumination face image reconstruction method and system |
| CN101976435A (en) * | 2010-10-07 | 2011-02-16 | 西安电子科技大学 | Combination learning super-resolution method based on dual constraint |
-
2013
- 2013-01-30 CN CN201310037260.1A patent/CN103632357B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101923721A (en) * | 2010-08-31 | 2010-12-22 | 汉王科技股份有限公司 | Non-illumination face image reconstruction method and system |
| CN101976435A (en) * | 2010-10-07 | 2011-02-16 | 西安电子科技大学 | Combination learning super-resolution method based on dual constraint |
Non-Patent Citations (2)
| Title |
|---|
| A face super-resolution method based on illumination invariant feature;Huang K. Ruimin Hu et al.;《Multimedia Technology(ICMT),2011 International Conference on》;20110728;5215-5218 * |
| 基于Log-WT的人脸图像超分辨率重建;乔建苹;《电子与信息学报》;20080630;1276-1280 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103632357A (en) | 2014-03-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bashir et al. | A comprehensive review of deep learning-based single image super-resolution | |
| US9460490B2 (en) | Image enhancement using semantic components | |
| Cai et al. | FCSR-GAN: Joint face completion and super-resolution via multi-task learning | |
| Roa'a et al. | Generation of high dynamic range for enhancing the panorama environment | |
| Li et al. | Single image snow removal via composition generative adversarial networks | |
| CN111062892A (en) | Single image rain removing method based on composite residual error network and deep supervision | |
| CN111696033B (en) | Real image super-resolution model and method based on angular point guided cascade hourglass network structure learning | |
| CN106920214A (en) | Spatial target images super resolution ratio reconstruction method | |
| CN103632357B (en) | A kind of image super-resolution Enhancement Method separated based on illumination | |
| Zheng et al. | T-net: Deep stacked scale-iteration network for image dehazing | |
| CN108492252A (en) | Face image super-resolution reconstruction method based on secondary reconstruction | |
| CN116012255A (en) | Low-light image enhancement method for generating countermeasure network based on cyclic consistency | |
| Wang et al. | Paccdu: pyramid attention cross-convolutional dual unet for infrared and visible image fusion | |
| Liang et al. | Deep convolution neural networks for automatic eyeglasses removal | |
| CN117011357A (en) | Human body depth estimation method and system based on 3D motion flow and normal map constraint | |
| CN116311384A (en) | Cross-modal pedestrian re-recognition method and device based on intermediate mode and characterization learning | |
| Pang et al. | Infrared and visible image fusion based on double fluid pyramids and multi-scale gradient residual block | |
| Huang et al. | Single image super-resolution reconstruction of enhanced loss function with multi-gpu training | |
| Zhang et al. | Face deblurring based on separable normalization and adaptive denormalization | |
| CN107463864A (en) | A kind of face identification system of binding sequence image super-resolution rebuilding | |
| Yue et al. | Sift-based image super-resolution | |
| CN117612266B (en) | Cross-resolution pedestrian re-identification method based on multi-scale image and feature layer alignment | |
| Liu et al. | Super-resolution reconstruction of thermal imaging of power equipment based on Generative Adversarial Network with Channel Filtering | |
| Zhang et al. | Temporal Squeeze-and-Excitation Networks for Skeleton-Based Action Recognition | |
| CN113487487B (en) | Super-resolution reconstruction method and system for heterogeneous stereo image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |