CN108986027A - Depth image super-resolution reconstruction method based on improved joint trilateral filter - Google Patents
Depth image super-resolution reconstruction method based on improved joint trilateral filter Download PDFInfo
- Publication number
- CN108986027A CN108986027A CN201810668468.6A CN201810668468A CN108986027A CN 108986027 A CN108986027 A CN 108986027A CN 201810668468 A CN201810668468 A CN 201810668468A CN 108986027 A CN108986027 A CN 108986027A
- Authority
- CN
- China
- Prior art keywords
- image
- edge
- interpolation
- resolution
- low resolution
- 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
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000012360 testing method Methods 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims abstract description 9
- 238000000605 extraction Methods 0.000 claims abstract description 7
- 230000003321 amplification Effects 0.000 claims abstract description 4
- 238000001914 filtration Methods 0.000 claims abstract description 4
- 238000003199 nucleic acid amplification method Methods 0.000 claims abstract description 4
- 239000012141 concentrate Substances 0.000 claims abstract description 3
- 239000000284 extract Substances 0.000 claims description 4
- 229910002056 binary alloy Inorganic materials 0.000 claims description 2
- 230000002708 enhancing effect Effects 0.000 claims description 2
- BTCSSZJGUNDROE-UHFFFAOYSA-N gamma-aminobutyric acid Chemical compound NCCCC(O)=O BTCSSZJGUNDROE-UHFFFAOYSA-N 0.000 claims 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 claims 1
- 239000010931 gold Substances 0.000 claims 1
- 229910052737 gold Inorganic materials 0.000 claims 1
- 230000000694 effects Effects 0.000 description 9
- 238000002474 experimental method Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 3
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000004873 anchoring Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002969 morbid Effects 0.000 description 1
- 238000004080 punching Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual 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
- 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
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/14—Transformations for image registration, e.g. adjusting or mapping for alignment of images
-
- 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
-
- 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
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
Depth image super-resolution reconstruction method based on improved joint trilateral filter.It include: with different decimation factors using bicubic interpolation operator to low resolution test image interpolation amplification, the edge image of image and low resolution chart picture after extracting interpolation respectively obtains edge image pyramid；Image block, composing training data set are extracted from image pyramid；Learnt using the image block that K-SVD algorithm concentrates training data, complete dictionary is obtained；Low resolution test image is amplified to target size using bicubic interpolation algorithm, is smoothed using jagged edges of the impact filtering to interpolation image, and the edge of the image after extraction process；Rarefaction representation is carried out by treated the edge image of the atom pair in dictionary, obtains the edge image of high quality；Under high quality margin guide, the three side filter coefficients of joint that low resolution test image improves are rebuild, high-resolution depth graph picture is obtained.
Description
Technical field
The invention belongs to computer vision and field of image processing more particularly to a kind of sides of depth image Super-resolution Reconstruction
Method.
Background technique
Depth image is mainly used for recording the object in scene to the distance between camera information, these information are to machine
The application and realization of people's navigation, augmented reality, human body attitude estimation etc. are crucial.In recent years, some depth cameras
Extensive use is received by its low cost and real-time effectiveness, such as Kinect and PMD (Photonic Mixer
Device) camera.But due to the influence of the limitation of depth camera internal hardware system and external environment, cause to directly acquire
Depth image limited resolution, so that it cannot meet the application of some aspects.If being improved by improving hardware system
Image resolution ratio, higher cost and is difficult to realize, therefore the side of the depth image Super-resolution Reconstruction using signal processing technology
Method is come into being.
The main purpose of depth image super resolution technology is to improve the resolution ratio of image, however the problem is that a morbid state is asked
Topic, the image of a width low resolution may correspond to several high-resolution images.So at this moment obtaining some prior informations in advance
Super-resolution Reconstruction process for depth image is considerable.According to the difference in prior information source, current depth image
The method of Super-resolution Reconstruction can be mainly divided into two classes: depth image ultra-resolution method based on learn-by-example and be based on cromogram
As the depth image ultra-resolution method of guidance.Depth image ultra-resolution method based on learn-by-example is by largely counting from external
According to some prior informations of focusing study, and then the reconstruction for instructing low resolution depth image.And drawing based on color image
The method led is with reference to the high-frequency information in the color image being registrated with low resolution depth map image height, and then guidance depth image
Super-resolution Reconstruction.
These two kinds of methods can successfully effectively improve the resolution ratio of depth image, but there is also certain deficiencies.Base
It needs to choose suitable external data collection by rule of thumb in the method for learn-by-example, and has stronger dependence to external data set
Property；Method based on color image guidance needs to obtain the color image with low resolution depth image height registration in advance, if
It can not be carried out without color image this method of height registration.So in order to adapt to the demand that various aspects are applied in real time, existing method
Present in some shortcomings, need to be solved, allow to the resolution ratio for simply and effectively improving depth image.
Summary of the invention
It is an object of the invention to for existing some depth image ultra-resolution methods to external data have it is stronger according to
The phenomenon that relying with reconstruction image edge sawtooth or artifact proposes that a kind of depth image based on improved joint trilateral filter is super
The method of resolved reconstruction.This method can be to avoid the dependence of external portion's data set, and can effectively keep marginal information
It is sharp.
It is super that the present invention provides a kind of depth image based on improved joint trilateral filter to solve above-mentioned technical problem
Resolved reconstruction method, method includes the following steps:
S1: using bicubic interpolation operator with different decimation factors to low resolution test image interpolation amplification, respectively
The edge image of image and low resolution chart picture after extracting interpolation, obtains an edge image pyramid；
S2: image block, composing training data set are extracted from image pyramid；
S3: being learnt using the image block that K-SVD algorithm concentrates training data, and then complete dictionary is obtained；
S4: low resolution test image is amplified to target size using bicubic interpolation algorithm, uses impact filtering pair
The jagged edges of interpolation image are handled, and the edge of the image after extraction process；
S5: rarefaction representation is carried out by treated the edge image of the atom pair in dictionary, obtains the edge graph of high quality
Picture；
S6: under high quality margin guide, carrying out three side filter coefficients of joint to low resolution test image and rebuild, thus
Obtain high-resolution depth graph picture.
Compared with prior art, the present invention achieves following advantageous effects:
1) present invention does not need to avoid the dependence to external data set to the selection process of external data set complexity；
2) present invention does not need the reference of the high-resolution color image of high registration；
3) present invention can not only keep the sharp of reconstruction image edge, but also can be in inhibition image effectively
Noise.
Detailed description of the invention
The content of claims to facilitate the understanding of the present invention and specific implementation process, some attached drawings are provided use
In showing reconstruction process of the invention and rebuild effect.Wherein:
Fig. 1 is the basic framework of the depth image ultra-resolution method based on improved joint trilateral filter in the present invention
Figure；
Fig. 2 is the edge image pyramid that the sparse reconstruction of edge image is used in the present invention；
Fig. 3 be the present invention with 4 kinds of classical ways to low resolution test image " bowling " super-resolution under four times of factors
Reconstructed results show；
Fig. 4 be the present invention with 4 kinds of classical ways to test image " dove " the Super-resolution Reconstruction knot under the four sampling factors
Fruit shows.
Specific embodiment
The present invention proposes the depth image super-resolution reconstruction method based on improved joint trilateral filter.In conjunction with the invention
To the basic framework figure that depth image is rebuild, Fig. 1 is described in detail specific implementation method of the invention:
S1: the test image D of one low resolution of inputl, interpolation is carried out with the decimation factor of i (i=2,3,4) to it and is put
Greatly, the image after obtaining interpolationExtract original low-resolution image DlWith the image after interpolationEdge imageWithTo constitute four layers of edge image pyramid T, edge pyramid such as Fig. 2 of building；
S2: image block, composing training data set { P are extracted from image pyramidk}g:
{Pk}g=RTg (1)
Wherein, R is linear extraction operator, for extracting image block；TgIt is pyramidal low j layers of edge image；PkIt indicates
K-th of image block of g layers of image pyramid extraction；
S3: using K-SVD algorithm to training dataset { Pk}jIn image block learnt, and then complete dictionary is obtained
A, specifically includes the following steps:
Wherein L is degree of rarefication constrained parameters, qkIt is to correspond to PkRarefaction representation coefficient matrix；
S4: using bicubic interpolation operator by the test image D of low resolutionl, interpolation amplification to target size uses punching
The edge of image after hitting filtering processing interpolation, to reduce sawtooth caused by interpolation, and extracts filtered image
Edge image obtains low quality edge image El；
S5: using the dictionary Α of training to edge image ElRarefaction representation is carried out, high quality edge Ε is constructedh, the step
In, using the sparsity of self-similarity and edge image between image block, pass through the atom pair edge image E in dictionarylIt carries out
Sparse reconstruction.Specifically includes the following steps:
Step 1: from edge image ΕlIn, extracting size isImage block(index that k is image block)；
Step 2: corresponding high quality graphic blockIt can be by the sparse linear combination table of atom in trained dictionary A
Show；
Step 3: final high quality edge image EhThe image block of middle extraction should be as close asBy following
Formula indicates:
Wherein, RkIt is that image block extracts operator, the size of image block is alsoHigh quality edge image EhIt can be with
It is acquired using least square method；
S6: using improved joint trilateral filter in high quality edge image EhGuidance under, to low resolution test
Image DlInterpolation reconstruction is carried out, to obtain desired high-resolution depth image Dh, the expression formula are as follows:
Wherein: Dh(p) pixel value of the position full resolution pricture p finally rebuild, k are indicatedpOne regularization factors, Ω table
Show the field window centered on pixel p, Dl(q ↓) indicates the low resolution test image D in inputlThe pixel at middle coordinate q ↓ place
Value, ΕhIndicating the high quality edge of reconstruct, p, ↓ and q ↓ respectively indicates pixel p and the coordinate of q, fs() indicates that variance is σsPicture
The distance between element Gaussian function, fg() indicates that variance is σgGradient information constraint function, WsIt is structural similarity index,
For enhancing the coherence of adjacent areas, fr() indicates a binary system indicator function, for whether differentiating two pixels
The same side.
For in formula 4, gradient information constraint function fgThe specific calculating process of () is as follows:
Assuming that in low resolution image DlThe coordinate position of middle pixel p is (i, j), is calculated first in the horizontal and vertical directions
The absolute value of First-order Gradient
If two pixels in adjacent edges are located at different depth planes, they may also have identical gradient distribution,
So having further calculated second order gradient to solve the problems, such as this, expression formula is as follows:
Then,WithIt will be by as two-dimensional Gaussian function fgThe input of () calculates between two pixels
Weight.
For in formula 4, structural similarity indexes WsCalculating process it is as follows:
Structural similarity SSIM is the important indicator for evaluating picture quality, which, which is used for weight adjacent areas, to reach
To preferable denoising effect.Structural similarity indexes WsBy mean function m (p, q), standard variance function σ (p, q) and structure phase
It is formed like property function s (p, q) three parts.Its expression formula difference is as follows:
Wherein, C1, C2And C3It is non-negative constant, the phenomenon that for avoiding denominator from being zero；μpAnd σpIt is centered on pixel p
The mean value and standard variance of all pixels in neighborhood；Similarly, μqAnd σqBe in the neighborhood centered on pixel q the mean value of pixel and
Standard variance, and pixel q is the pixel in field window centered on p；σpqIt is the covariance of two neighborhood windows.Finally,
Structural similarity indexes WsIt can be expressed as follows:
Ws=SSIM (p, q)=m (p, q)ασ(p,q)βs(p,q)γ (12)
Wherein, α, β and γ are weight factor.
The present invention can further illustrate the reconstruction effect of depth image by following experiment:
1, experiment condition:
1) experiment depth image used is from Middledury image set；
2) the low resolution test image in experiment is obtained by data set middle high-resolution image down sampling；
3) test based on programming platform be MATLAB2016a；
4) test computer used be configured to Intel (R) Xeon (R) CPU E5-2620v3@2.40H, 64.0GB RAM,
64 Win8 of operating system；
5) Y-PSNR (PSNR), structural similarity (SSIM), root-mean-square error (RMSE) and mistake are used in experiment
Index is objectively evaluated than 4 kinds of (PE) to evaluate the reconstructed results of image.
6) some parameters in experiment are to obtain the smallest RMSE value by many experiments to determine.
2, experiment content
Under the same conditions, four kinds of classical depth image super-resolution reconstruction methods be provided with method of the invention into
Row compares.These four methods include: Timofte【1】Et al. improved anchoring neighborhood regression algorithm, Kim【2】Et al. convolution
The method of neural network, Yang【3】Et al. the method for rarefaction representation, Zeyde【4】Et al. the method for rarefaction representation, Xie【5】Deng
The method that margin guide is rebuild based on Markov field of people.
The setting of method parameter is as shown in table 1 in the present invention:
1 RMSE value of table compares (4 times of reconstructions)
Wherein, n indicates to extract the square value of image block side length, and Ω indicates the field window when combining three side interpolation reconstructions
Size, σsAnd σgRespectively indicate Gaussian function fs() and fgThe variance yields of (), C1、C2And C3Respectively indicate structural similarity rope
Draw the non-negative constant in calculating, α, β and γ respectively indicate the weight of three functions in structural similarity index calculating.
Table 2-5 presents 4 kinds of comparative approach and the method for the present invention (Ours) and imitates in 4 kinds of reconstructions objectively evaluated in index
Fruit.The superiority and inferiority of the numerical value in comparison table for clarity, separately below in each table rebuild effect ranking before 2 numerical value
It is marked, the numerical value of overstriking body indicates optimal effectiveness, and the numerical result of underscore takes second place.
2 RMSE value of table compares (4 times of reconstructions)
3 SSIM value of table compares (4 times of reconstructions)
4 PSNR value of table compares (4 times of reconstructions)
5 PE value of table compares (4 times of reconstructions)
3, analysis of experimental results
Reading for clarity, we are marked to rebuilding effect ranking the first two numerical value in table.Black matrix overstriking
Numerical value be shown to be best reconstructed results, there is the numerical value of single underscore to take second place.It can from the numerical value in table 2 and table 4
Out, in the method compared, method of the invention is to the RMSE value and PSNR value of the reconstruction effect of depth image the nine of test
Opening can rank the first in image；Meanwhile it is also available from table 3 and table 5, in all test images of selection, this hair
It is bright rebuild effect SSIM and PE value can ranking the first two.
In order to assess the visual effect of image after the method for the present invention is rebuild, test image is provided in figs. 3 and 4
Image after original high-resolution image and the several method reconstruction of ' bowling ' and ' dove '.Wherein, Fig. 3,4 (a) be original
Full resolution pricture；Fig. 3,4 (b) be Timofe【1】The result that method is rebuild；Fig. 3,4 (c) be Kim【2】The result that method is rebuild；
Fig. 3,4 (d) be Zeyde【4】The result that method is rebuild；Fig. 3,4 (e) be Xie【5】The result that method is rebuild；Fig. 3,4 (f) be this hair
The result that bright method is rebuild；From these images, it can be observed that the image that the method for the present invention is rebuild not only can be to avoid generation
Fuzzy artifact, and help to reduce the sawtooth of adjacent edges.
Above said content is according to the detailed description made for the present invention of preferable embodiment, but it cannot be assumed that this hair
Bright specific implementation is only not limited to this.For being familiar with for person skilled in the art of the present invention, the present invention is not being departed from
Substitutions and modifications are made in the technical scope showed, and when purposes is identical with effect, should all be covered in protection of the invention
Within the scope of.
Bibliography
1.Timofte,R.,Smet,V.D.,Gool,L.V.:A+:Adjusted Anchored Neighborhood
Regression for Fast Super-Resolution.In:Asian Conference on Computer Vision,
pp.111-126.Singapore(2014).
2.Kim,J.,Kwon,L.J.,Mu,L.K.:Accurate image super-resolution using very
deep convolutional networks.In:Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition, pp.1646-1654.Las Vegas,NV,United States
(2016).Available in http://doi.ieeecomputersociety.org/10.1109/CVPR.2016.182
3.Yang,J.,Wright,J.,Huang,T.S.:Image super-resolution via sparse
representation.IEEE transactions on image processing 19(11),2861-2873(2010).
4.Zeyde,R.,Elad,M.,Protter,M.:On single image scale-up using sparse-
representations.In: International conference on curves and surfaces,711-
730.Springer Berlin Heidelberg(2010).
5.Xie,J.,Feris,R.S.,Sun,M.T.:Edge-guided single depth image super
resolution.IEEE Transactions on Image Processing 25(1),428-438(2016)。
Claims (4)
1. the depth image super-resolution reconstruction method based on improved joint trilateral filter, which is characterized in that this method includes
Following steps:
S1: low resolution test image interpolation amplification is extracted respectively with different decimation factors using bicubic interpolation operator
The edge image of image and low resolution chart picture after interpolation, obtains an edge image pyramid；
S2: image block, composing training data set are extracted from image pyramid；
S3: being learnt using the image block that K-SVD algorithm concentrates training data, and then complete dictionary is obtained；
S4: low resolution test image is amplified to target size using bicubic interpolation algorithm, using impact filtering to interpolation
The jagged edges of image are smoothed, and the edge of the image after extraction process；
S5: rarefaction representation is carried out by treated the edge image of the atom pair in dictionary, obtains the edge image of high quality；
S6: under high quality margin guide, interpolation is carried out to low resolution test image using improved joint trilateral filter
It rebuilds, to obtain high-resolution depth graph picture.
2. the depth image super-resolution reconstruction method as described in claim 1 based on improved joint trilateral filter, special
Sign is, in step sl, obtains edge image pyramid according to following steps:
1) with decimation factor i (i=2,3,4) to low resolution chart as DlCarry out interpolation, the image after obtaining interpolation
2) low resolution chart is extracted respectively as DlAnd imageEdge imageWithConstitute four layers of edge image gold word
Tower, wherein l indicates that picture is low-resolution image, and i indicates the different decimation factor of image.
3. the depth image super-resolution reconstruction method as described in claim 1 based on improved joint trilateral filter, special
Sign is, in step s 2, using the linear operator that extracts to the image progress image block { P in image pyramidk}jExtraction；Its
In, k indicates that the index of image block, j indicate the pyramidal number of plies.
4. the depth image super-resolution reconstruction method as described in claim 1 based on improved joint trilateral filter, special
Sign is, the equation of improved joint trilateral filter are as follows:
Wherein: Dh(p) pixel value of the position full resolution pricture p finally rebuild, k are indicatedpOne regularization factors, Ω indicate with
Field window centered on pixel p, Dl(q ↓) indicates the low resolution test image D in inputlThe pixel value at middle coordinate q ↓ place,
ΕhIndicating the high quality edge of reconstruct, p, ↓ and q ↓ respectively indicates pixel p and the coordinate of q, fs() indicates the distance between pixel
Gaussian function, fg() indicates gradient information constraint function, WsIt is structural similarity index, for enhancing adjacent areas's
Coherence, fr() indicate a binary system indicator function, for differentiate two pixels whether the same side.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810668468.6A CN108986027A (en) | 2018-06-26 | 2018-06-26 | Depth image super-resolution reconstruction method based on improved joint trilateral filter |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810668468.6A CN108986027A (en) | 2018-06-26 | 2018-06-26 | Depth image super-resolution reconstruction method based on improved joint trilateral filter |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108986027A true CN108986027A (en) | 2018-12-11 |
Family
ID=64538331
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810668468.6A Pending CN108986027A (en) | 2018-06-26 | 2018-06-26 | Depth image super-resolution reconstruction method based on improved joint trilateral filter |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108986027A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139918A (en) * | 2021-04-23 | 2021-07-20 | 大连大学 | Image reconstruction method based on decision-making gray wolf optimization dictionary learning |
| CN114459414A (en) * | 2021-12-23 | 2022-05-10 | 宜昌测试技术研究所 | Depth detection method for semi-submersible navigation body |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106408550A (en) * | 2016-09-22 | 2017-02-15 | 天津工业大学 | Improved self-adaptive multi-dictionary learning image super-resolution reconstruction method |
| CN107169928A (en) * | 2017-05-12 | 2017-09-15 | 武汉华大联创智能科技有限公司 | A kind of human face super-resolution algorithm for reconstructing learnt based on deep layer Linear Mapping |
| CN107767357A (en) * | 2017-09-14 | 2018-03-06 | 北京工业大学 | A kind of depth image super-resolution method based on multi-direction dictionary |
-
2018
- 2018-06-26 CN CN201810668468.6A patent/CN108986027A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106408550A (en) * | 2016-09-22 | 2017-02-15 | 天津工业大学 | Improved self-adaptive multi-dictionary learning image super-resolution reconstruction method |
| CN107169928A (en) * | 2017-05-12 | 2017-09-15 | 武汉华大联创智能科技有限公司 | A kind of human face super-resolution algorithm for reconstructing learnt based on deep layer Linear Mapping |
| CN107767357A (en) * | 2017-09-14 | 2018-03-06 | 北京工业大学 | A kind of depth image super-resolution method based on multi-direction dictionary |
Non-Patent Citations (2)
| Title |
|---|
| JUN XIE 等: "Edge-Guided Single Depth Image Super Resolution", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| 王建新 等: "残差字典学习的快速图像超分辨率算法", 《计算机科学与探索》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139918A (en) * | 2021-04-23 | 2021-07-20 | 大连大学 | Image reconstruction method based on decision-making gray wolf optimization dictionary learning |
| CN113139918B (en) * | 2021-04-23 | 2023-11-10 | 大连大学 | Image reconstruction method based on decision-making gray wolf optimization dictionary learning |
| CN114459414A (en) * | 2021-12-23 | 2022-05-10 | 宜昌测试技术研究所 | Depth detection method for semi-submersible navigation body |
| CN114459414B (en) * | 2021-12-23 | 2023-12-19 | 宜昌测试技术研究所 | Depth detection method for semi-submersible vehicle |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ma et al. | Infrared and visible image fusion via detail preserving adversarial learning | |
| CN110119780B (en) | Hyper-spectral image super-resolution reconstruction method based on generation countermeasure network | |
| CN107154023B (en) | Based on the face super-resolution reconstruction method for generating confrontation network and sub-pix convolution | |
| CN106952228B (en) | Super-resolution reconstruction method of single image based on image non-local self-similarity | |
| CN106339998B (en) | Multi-focus image fusing method based on contrast pyramid transformation | |
| CN105741252B (en) | Video image grade reconstruction method based on rarefaction representation and dictionary learning | |
| Kumar et al. | Convolutional neural networks for wavelet domain super resolution | |
| Yuan et al. | Regional spatially adaptive total variation super-resolution with spatial information filtering and clustering | |
| CN108133456A (en) | Face super-resolution reconstruction method, reconstructing apparatus and computer system | |
| DE102020214863A1 (en) | SELF-MONITORED PROCEDURE AND SYSTEM FOR DEPTH ESTIMATION | |
| CN106127688B (en) | A kind of super-resolution image reconstruction method and its system | |
| CN105960657A (en) | Face hallucination using convolutional neural networks | |
| CN103208102A (en) | Remote sensing image fusion method based on sparse representation | |
| CN105046672A (en) | Method for image super-resolution reconstruction | |
| CN102354397A (en) | Method for reconstructing human facial image super-resolution based on similarity of facial characteristic organs | |
| CN105488759B (en) | A kind of image super-resolution rebuilding method based on local regression model | |
| CN102982520B (en) | Robustness face super-resolution processing method based on contour inspection | |
| CN108765280A (en) | A kind of high spectrum image spatial resolution enhancement method | |
| CN104021523B (en) | A kind of method of the image super-resolution amplification based on marginal classification | |
| Gai et al. | Multi-focus image fusion method based on two stage of convolutional neural network | |
| Xiao et al. | A dual-UNet with multistage details injection for hyperspectral image fusion | |
| CN104252703B (en) | Wavelet preprocessing and sparse representation-based satellite remote sensing image super-resolution reconstruction method | |
| CN107767357B (en) | Depth image super-resolution method based on multi-direction dictionary | |
| CN106169174A (en) | A kind of image magnification method | |
| CN108986027A (en) | Depth image super-resolution reconstruction method based on improved joint trilateral filter |
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 |