CN109785235A - Compressed sensing based infrared pedestrian image super resolution ratio reconstruction method and system - Google Patents
Compressed sensing based infrared pedestrian image super resolution ratio reconstruction method and system Download PDFInfo
- Publication number
- CN109785235A CN109785235A CN201811644529.1A CN201811644529A CN109785235A CN 109785235 A CN109785235 A CN 109785235A CN 201811644529 A CN201811644529 A CN 201811644529A CN 109785235 A CN109785235 A CN 109785235A
- Authority
- CN
- China
- Prior art keywords
- resolution
- image
- low
- dictionary
- block
- 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 47
- 238000012549 training Methods 0.000 claims abstract description 57
- 238000012512 characterization method Methods 0.000 claims abstract description 18
- 230000003321 amplification Effects 0.000 claims description 7
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000009499 grossing Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 239000000284 extract Substances 0.000 abstract description 2
- 238000003331 infrared imaging Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 235000007926 Craterellus fallax Nutrition 0.000 description 1
- 240000007175 Datura inoxia Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000005215 recombination Methods 0.000 description 1
- 230000006798 recombination Effects 0.000 description 1
Landscapes
- Image Processing (AREA)
- Transforming Light Signals Into Electric Signals (AREA)
- Closed-Circuit Television Systems (AREA)
Abstract
The invention discloses a kind of compressed sensing based infrared pedestrian image super resolution ratio reconstruction method and systems, method includes: that two infrared detectors of high-resolution and low resolution are placed adjacent in face of Same Scene, acquire high-low resolution two types image under same scene, low-resolution image is amplified to and full resolution pricture same size using linear interpolation, then image block is divided the image into, constructs the training sample set of high-low resolution dictionary respectively；Extract characteristics of image, the feature of high-resolution and low-resolution image block is trained using identical sparse coding algorithm, high-resolution and low-resolution dictionary is obtained, so that high-definition picture block and low-resolution image the block rarefaction representation having the same under the characterization of respective corresponding dictionary；Feature is extracted to the infrared pedestrian image of the low resolution of input, the rarefaction representation of image is obtained under low-resolution dictionary characterization；The sparse representation and high-resolution dictionary of the image obtained by step are rebuild, and high-definition picture is obtained.
Description
Technical field
The present invention relates to image super-resolution rebuilding fields, more particularly to a kind of compressed sensing based infrared pedestrian image
Super resolution ratio reconstruction method and system.
Background technique
Infrared imagery technique has the characteristics that anti-interference and recognition capability is strong, in video monitoring, medicine detection and industry inspection
There is extensive application in survey field.It is most of to be adopted using the infrared detector of partial array in current infrared imaging application
Collect image.However, with the continuous extension of application range, the collected low resolution infrared image of partial array infrared imaging device
It is difficult to meet growing application demand.Therefore, the image resolution ratio for improving infrared imaging has highly important reality meaning
Justice.Infrared detector using large area array is most direct effective method, and due to the limitation of technical level, domestic large area array is visited
Survey device performance and it is external there are a certain distance, external is often at high price, the considerations of for equipment cost, it is difficult to extensively
Using；Improving the resolution ratio of image by the method for image procossing is another means, can be passed through under the conditions of existing equipment
The mode of software algorithm promotes picture quality.Past 10 years, compressive sensing theory achieved perhaps in the application of field of image processing
The achievement more attracted attention.The Super-Resolution of Images Based for using for reference the advanced theory of compressed sensing, can significantly improve the quality of image, have
There is highly important realistic meaning.
The super-resolution rebuilding technology of image can be divided into three classes according to principle division: based on interpolation, based on reconstructing and be based on
The method of study.Method based on interpolation realizes that simply arithmetic speed is fast, but it is difficult to the high frequency for restoring to lose when imaging letter
Breath, thus image quality is poor；Method based on reconstruct be to several low-resolution images of same scene in spatial domain or
Transform domain reconstructs high-definition picture, such method there are certain requirements Image Acquisition, and application has must limitation；
And the Super-resolution Reconstruction algorithm based on study is to obtain reflecting between low-resolution image and high-definition picture by training set
Relationship is penetrated, under conditions of known low explanation image, solves optimal value, i.e. high-definition picture using the mapping relations.
Compressive sensing theory gives the method that original high dimensional signal is recovered from low-dimensional signal high precision.It will compression
Perception combined with Image Super-resolution can with high quality, efficiently from low explanation image reconstruction go out high-definition picture, realization
Using existing infrared imaging device, the target of high-resolution pedestrian image is obtained.
Summary of the invention
The technical problem to be solved in the present invention is that the problem low for existing infrared imaging device imaging resolution, provides
A kind of compressed sensing based super resolution ratio reconstruction method, the infrared pedestrian image of collected low resolution, which is rebuild, becomes high score
Distinguish image.
The technical solution adopted by the present invention to solve the technical problems is:
A kind of compressed sensing based infrared pedestrian image super resolution ratio reconstruction method is provided, comprising the following steps:
Two infrared detectors of high-resolution and low resolution are placed adjacent by the collection and pretreatment of S1, training sample
In face of Same Scene, high-low resolution two types image under same scene is acquired, low-resolution image is put using linear interpolation
It is big then to divide the image into image block extremely with full resolution pricture same size, the training of high-low resolution dictionary is constructed respectively
Sample set；
The training of S2, high-resolution dictionary and low-resolution dictionary are extracted characteristics of image, are calculated using identical sparse coding
Method is trained the feature of high-resolution and low-resolution image block, high-resolution and low-resolution dictionary is obtained, so that high-definition picture block
With low-resolution image block under the characterization of respective corresponding dictionary rarefaction representation having the same；
S3, feature is extracted to the infrared pedestrian image of the low resolution of input, obtains image under low-resolution dictionary characterization
Rarefaction representation；
S4, it is rebuild by the obtained sparse representation of step S3 and high-resolution dictionary, obtains high-definition picture.
Above-mentioned technical proposal is connect, the high-resolution is 640 × 512, and low resolution is 320 × 256.
Meet above-mentioned technical proposal, step S1 specifically:
S101, two infrared detectors are set up, be placed adjacent in face of Same Scene, one is high-resolution detector, separately
One is low resolution detector, while acquiring video image；
S102, using linear interpolation algorithm, the video image that low resolution infrared detector acquires is extended to high-resolution
The video image of rate；
In S103, two groups of video images, the picture frame of synchronization under same scene is found out, forms image pair；For every
One image pair, the image block for carrying out 9 × 9 to two sub-pictures respectively operate；High-definition picture block forms high-resolution dictionary
Training sample set, and low-resolution image block forms low-resolution dictionary training sample set.
Above-mentioned technical proposal is connect, in step S2 specifically:
S201, high-resolution dictionary DhWith low-resolution dictionary DlIt is obtained respectively by following formula training:
In formula, XhAnd YlIt is high-resolution sample set and low resolution sample set respectively；
S202, during training dictionary, using identical sparse coding algorithm come to the image in S103 to carry out it is same
Shi Xunlian, this process are described by following formula:
Wherein, N and M is respectively the dimension of high-resolution and low-resolution image block vector in sample database establishment process.
Meet above-mentioned technical proposal, step S3 specifically:
S301, by low-resolution image I1Then interpolation amplification will to desired high-definition picture same size
Amplified image carries out piecemeal and constitutes low-resolution image block；
S302, acquisition low-resolution image block solve its it is low explanation dictionary characterization under rarefaction representation vector α, this is dilute
Dredging indicates that vector is the corresponding rarefaction representation vector of corresponding high-definition picture block.
Meet above-mentioned technical proposal, step S4 specifically:
The high-resolution dictionary D that S401, image block rarefaction representation vector α and the S201 training obtained by S302 obtainh, weight
Structure goes out high-definition picture block, calculation formula are as follows: Dhα；
S402, calculated high-definition picture block, are spliced into complete image；And to the side between image block
Do the smoothing processing of image in boundary.
The present invention also provides a kind of compressed sensing based infrared pedestrian image super-resolution rebuilding systems, comprising:
The collection and preprocessing module of training sample, for high-resolution and two infrared detectors of low resolution is adjacent
Placed side acquires high-low resolution two types image under same scene to Same Scene, and low-resolution image is inserted using linear
Value be amplified to full resolution pricture same size, then divide the image into image block, construct high-low resolution dictionary respectively
Training sample set；
The training module of high-resolution dictionary and low-resolution dictionary, for extracting characteristics of image, using identical sparse
Encryption algorithm is trained the feature of high-resolution and low-resolution image block, high-resolution and low-resolution dictionary is obtained, so that high-resolution
Image block and low-resolution image the block rarefaction representation having the same under the characterization of respective corresponding dictionary；
Rarefaction representation module, for extracting feature to the infrared pedestrian image of the low resolution of input, in low-resolution dictionary
The lower rarefaction representation for obtaining image of characterization；
Module is rebuild, for being rebuild by the rarefaction representation and high-resolution dictionary of obtained image, obtains high score
Resolution image.
Above-mentioned technical proposal is connect, the collection of the training sample and preprocessing module are specifically used for:
Video image is acquired simultaneously by two infrared detectors of erection, and two infrared detectors are placed adjacent in face of same
One scene, one is high-resolution detector, and another is low resolution detector；
Using linear interpolation algorithm, the video image that low resolution infrared detector acquires is extended to high-resolution view
Frequency image；
In two groups of video images, the picture frame of synchronization under same scene is found out, forms image pair；For each figure
As right, the image block for carrying out 9 × 9 to two sub-pictures respectively is operated；High-definition picture block forms high-resolution dictionary training sample
This collection, and low-resolution image block forms low-resolution dictionary training sample set.
Connect above-mentioned technical proposal, the training module of the high-resolution dictionary and low-resolution dictionary is specifically used for:
High-resolution dictionary D is obtained by following formula traininghWith low-resolution dictionary Dl:
In formula, XhAnd YlIt is high-resolution sample set and low resolution sample set respectively；
During training dictionary, to progress while being trained using identical sparse coding algorithm come the image to acquisition,
This process is described by following formula:
Wherein, N and M is respectively the dimension of high-resolution and low-resolution image block vector in sample database establishment process.
Above-mentioned technical proposal is connect, the rarefaction representation module is specifically used for:
By low-resolution image I1Interpolation amplification, then will be after amplification to desired high-definition picture same size
Image carry out piecemeal constitute low-resolution image block；
The low-resolution image block of acquisition solves its rarefaction representation vector α under low explanation dictionary characterization, the sparse table
Show vector for the corresponding rarefaction representation vector of corresponding high-definition picture block；
The reconstruction module is specifically used for:
The high-resolution dictionary D obtained by obtained image block rarefaction representation vector α and trainingh, reconstruct high-resolution
Image block, calculation formula are as follows: Dhα；
Calculated high-definition picture block, is spliced into complete image；And the boundary between image block is done
The smoothing processing of image.
The beneficial effect comprise that: compressive sensing theory is integrated to the processing of infrared pedestrian image by the present invention
In, realize the super-resolution recombination function of image.Based on the theory, acquires the infrared pedestrian image of high-resolution and low-resolution and construct image block
Sample set is for training high-resolution and low-resolution dictionary, same target rarefaction representation system having the same under high-resolution and low-resolution dictionary
Number.Therefore, using trained dictionary, its expression coefficient under low explanation dictionary, then benefit are solved by low-resolution image
Corresponding full resolution pricture can be reconstructed with high-resolution dictionary.The image detail that the present invention is rebuild is compared with horn of plenty, computational efficiency
Height can be used as Image post-processing algorithm stable operation in current infrared monitoring equipment, promote current device acquired image
Resolution ratio.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples, in attached drawing:
Fig. 1 is the infrared pedestrian image super resolution ratio reconstruction method flow chart the present invention is based on compressed sensing；
Fig. 2 is the exemplary diagram of full resolution pricture block in the embodiment of the present invention；
Fig. 3 is the compressed sensing based infrared pedestrian image super-resolution rebuilding system structure signal of the embodiment of the present invention
Figure.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention.
Same scene is acquired first the present invention is based on the infrared pedestrian image super resolution ratio reconstruction method of compressed sensing and is schemed
Picture, using high-resolution and low-resolution image to building high-resolution and low-resolution sample set, for training high-resolution and low-resolution dictionary.Training
In image will be subjected to piecemeal first, image block is detected using algorithm of target detection, guarantee image block in detected
Whole or partial pedestrian target.Table of the image block of corresponding position in high-resolution and low-resolution dictionary in high-resolution and low-resolution image
Under sign, rarefaction representation coefficient having the same.Therefore, high-resolution reconstruction process is that low explanation image is utilized low resolution
Dictionary solves rarefaction representation coefficient, reconstructs full resolution pricture by the coefficient and high-resolution dictionary.
Present example provides a kind of compressed sensing based infrared pedestrian image super resolution ratio reconstruction method, method stream
Journey figure is shown in that Fig. 1, detailed description see below:
S01, the collection and pretreatment of training sample, by high-resolution (640 × 512) and low resolution (320 × 256) two
A infrared detector is placed adjacent in face of Same Scene, acquires high-low resolution two types image under same scene, low resolution
Rate image using linear interpolation be amplified to full resolution pricture same size (640 × 512), then divide the image into image block
The training sample set of high-low resolution dictionary is constructed respectively；
Further, in step S01 of the invention, the collection of used training sample and pretreated method are specific
Are as follows:
S101 sets up two infrared detectors, is placed adjacent in face of Same Scene, and one is high-resolution (640 × 512)
Detector, another is low resolution (320 × 256) detector, while acquiring infrared pedestrian's video image.
The video image that low resolution infrared detector acquires is extended to resolution ratio using linear interpolation algorithm by S102
For 640 × 512 video image.
S103 in two groups of video images, filters out the IR video stream comprising pedestrian, finds out under same scene with for the moment
The picture frame at quarter forms image pair.Then low-resolution image and full resolution pricture are carried out by image registration using SIFT operator.
For each image pair, the image block for carrying out 9 × 9 to two sub-pictures respectively is operated, by empirical analysis, the method for partition
It can guarantee that each image block under full resolution pricture and low-resolution image includes the information of pedestrian.High resolution graphics
As block composition high-resolution dictionary training sample set, and low-resolution image block forms low-resolution dictionary training sample set.Training
Sample set quantity is bigger, and the dictionary of building is just more accurate, in addition, training sample also need to collect extensively as far as possible it is various types of
The infrared pedestrian image of type, similarity is smaller between sample, and sample is abundanter, and the application range of algorithm can be more wide.
S02, the training of high-resolution dictionary and low-resolution dictionary, first extraction characteristics of image, using identical sparse
Encryption algorithm is trained the feature of high-resolution and low-resolution image block, high-resolution and low-resolution dictionary is obtained, so that high-resolution
Image block and low-resolution image the block rarefaction representation having the same under the characterization of respective corresponding dictionary；
Further, in step S02 of the invention, the training method of high-resolution dictionary and low-resolution dictionary is specific
Are as follows:
S201 high-resolution dictionary DhWith low-resolution dictionary DlIt is obtained respectively by following formula training:
In formula, XhAnd YlIt is high-resolution sample set and low resolution sample set respectively, λ is regularization parameter, ‖ ‖1Table
Show 1 rank norm of matrix.
S202 is during training dictionary, in order to make the image of same target different resolution have consistent rarefaction representation
It is sparse, it needs using identical sparse coding algorithm come to the image in S103, to progress while training, this process is by following
Formula description:
Wherein, N and M is respectively the dimension of high-resolution and low-resolution image block vector in sample database establishment process.
Constructing dictionary, we use K rank singular value decomposition algorithm, i.e. K-SVD algorithm.The algorithm of K-SVD includes three steps
It is rapid: random initializtion dictionary D；Fixed dictionary, seeks the sparse coding of each sample；Dictionary is updated by column, and is updated corresponding
Non-zero code.
S03 extracts feature to the low-resolution image of input, the sparse table of image is obtained under low-resolution dictionary characterization
Show；
Further, in step S03 of the invention low-resolution image rarefaction representation method specifically:
S301, by low-resolution image I1Interpolation amplification obtains figure to desired high-definition picture same size
As Xl, amplified image is then subjected to piecemeal and constitutes low-resolution image block { xl}。
S302, the low-resolution image block of acquisition solve its it is low explanation dictionary characterization under rarefaction representation vector α, this to
Amount is the corresponding rarefaction representation vector of corresponding high-definition picture block.
Rarefaction representation coefficient α is solved by following formula:
F is high-pass filter, for extracting the single order and second order gradient of image block.Solving above-mentioned formula can be obtained image
Block xlIn low explanation dictionary DlUnder rarefaction representation coefficient α.
S04, the sparse representation and high-resolution dictionary that high-definition picture is obtained by back are rebuild.
Further, the method for step S4 middle high-resolution image reconstruction of the invention specifically:
S401, the high-resolution dictionary D that image block rarefaction representation vector α and the S201 training obtained by S302 obtainsh, weight
Structure goes out high-definition picture block, calculation formula are as follows:
xh=Dhα
S402, calculated high-definition picture block { xh, it is spliced into complete image.In addition, in order to enable
Keep continuously and compatible between image block, the boundary between image block also needs to do the smoothing processing of image.
Method in order to realize above-described embodiment, the embodiment of the invention also provides compressed sensing based infrared pedestrians to scheme
As super-resolution rebuilding system, as shown in figure 3, specifically including:
The collection and preprocessing module of training sample, for high-resolution and two infrared detectors of low resolution is adjacent
Placed side acquires high-low resolution two types image under same scene to Same Scene, and low-resolution image is inserted using linear
Value be amplified to full resolution pricture same size, then divide the image into image block, construct high-low resolution dictionary respectively
Training sample set；
The training module of high-resolution dictionary and low-resolution dictionary, for extracting characteristics of image, using identical sparse
Encryption algorithm is trained the feature of high-resolution and low-resolution image block, high-resolution and low-resolution dictionary is obtained, so that high-resolution
Image block and low-resolution image the block rarefaction representation having the same under the characterization of respective corresponding dictionary；
Rarefaction representation module, for extracting feature to the infrared pedestrian image of the low resolution of input, in low-resolution dictionary
The lower rarefaction representation for obtaining image of characterization；
Module is rebuild, is rebuild for passing through obtained rarefaction representation and high-resolution dictionary, obtains high resolution graphics
Picture.
The specific implementation process of modules is as the method and step of above-described embodiment, and this will not be repeated here.
It should be understood that for those of ordinary skills, it can be modified or changed according to the above description,
And all these modifications and variations should all belong to the protection domain of appended claims of the present invention.
Claims (10)
1. a kind of compressed sensing based infrared pedestrian image super resolution ratio reconstruction method, which comprises the following steps:
Two infrared detectors of high-resolution and low resolution are placed adjacent and face by the collection and pretreatment of S1, training sample
Same Scene, acquires high-low resolution two types image under same scene, and low-resolution image is amplified to using linear interpolation
With full resolution pricture same size, image block is then divided the image into, constructs the training sample of high-low resolution dictionary respectively
Collection；
Characteristics of image is extracted in the training of S2, high-resolution dictionary and low-resolution dictionary, is come using identical sparse coding algorithm
The feature of high-resolution and low-resolution image block is trained, high-resolution and low-resolution dictionary is obtained, so that high-definition picture block and low
Image in different resolution block rarefaction representation having the same under the characterization of respective corresponding dictionary；
S3, feature is extracted to the infrared pedestrian image of the low resolution of input, obtains the dilute of image under low-resolution dictionary characterization
Dredging indicates；
S4, it is rebuild by the sparse representation and high-resolution dictionary of the obtained image of step S3, obtains high-definition picture.
2. compressed sensing based infrared pedestrian image super resolution ratio reconstruction method according to claim 1, feature exist
In the high-resolution is 640 × 512, and low resolution is 320 × 256.
3. compressed sensing based infrared pedestrian image super resolution ratio reconstruction method according to claim 1, feature exist
In step S1 specifically:
S101, two infrared detectors are set up, be placed adjacent in face of Same Scene, one is high-resolution detector, Ling Yitai
For low resolution detector, while acquiring video image；
S102, using linear interpolation algorithm, the video image that low resolution infrared detector acquires is extended to high-resolution
Video image；
In S103, two groups of video images, the picture frame of synchronization under same scene is found out, forms image pair；For each
Image pair, the image block for carrying out 9 × 9 to two sub-pictures respectively operate；High-definition picture block forms the training of high-resolution dictionary
Sample set, and low-resolution image block forms low-resolution dictionary training sample set.
4. compressed sensing based infrared pedestrian image super resolution ratio reconstruction method according to claim 3, feature exist
In in step S2 specifically:
S201, high-resolution dictionary DhWith low-resolution dictionary DlIt is obtained respectively by following formula training:
In formula, XhAnd YlIt is high-resolution sample set and low resolution sample set respectively；
S202, during training dictionary, using identical sparse coding algorithm come to the image in S103 to carry out and meanwhile instruct
Practice, this process is described by following formula:
Wherein, N and M is respectively the dimension of high-resolution and low-resolution image block vector in sample database establishment process.
5. compressed sensing based infrared pedestrian image super resolution ratio reconstruction method according to claim 4, feature exist
In step S3 specifically:
S301, by low-resolution image I1Interpolation amplification, then will be after amplification to desired high-definition picture same size
Image carry out piecemeal constitute low-resolution image block；
S302, acquisition low-resolution image block solve its it is low explanation dictionary characterization under rarefaction representation vector α, the sparse table
Show vector for the corresponding rarefaction representation vector of corresponding high-definition picture block.
6. compressed sensing based infrared pedestrian image super resolution ratio reconstruction method according to claim 5, feature exist
In step S4 specifically:
The high-resolution dictionary D that S401, image block rarefaction representation vector α and the S201 training obtained by S302 obtainh, reconstruct
High-definition picture block, calculation formula are as follows: Dhα；
S402, calculated high-definition picture block, are spliced into complete image；And the boundary between image block is done
The smoothing processing of image.
7. a kind of compressed sensing based infrared pedestrian image super-resolution rebuilding system characterized by comprising
The collection and preprocessing module of training sample, for being placed adjacent two infrared detectors of high-resolution and low resolution
In face of Same Scene, high-low resolution two types image under same scene is acquired, low-resolution image is put using linear interpolation
It is big then to divide the image into image block extremely with full resolution pricture same size, the training of high-low resolution dictionary is constructed respectively
Sample set；
The training module of high-resolution dictionary and low-resolution dictionary uses identical sparse coding for extracting characteristics of image
Algorithm is trained the feature of high-resolution and low-resolution image block, high-resolution and low-resolution dictionary is obtained, so that high-definition picture
Block and low-resolution image the block rarefaction representation having the same under the characterization of respective corresponding dictionary；
Rarefaction representation module is characterized for extracting feature to the infrared pedestrian image of the low resolution of input in low-resolution dictionary
The lower rarefaction representation for obtaining image；
Module is rebuild, for being rebuild by the rarefaction representation and high-resolution dictionary of obtained image, obtains high-resolution
Image.
8. according to the compressed sensing based infrared pedestrian image super-resolution rebuilding system of claim 7, which is characterized in that described
The collection of training sample and preprocessing module are specifically used for:
Video image is acquired simultaneously by two infrared detectors of erection, two infrared detectors are placed adjacent in face of same field
Scape, one is high-resolution detector, and another is low resolution detector；
Using linear interpolation algorithm, the video image that low resolution infrared detector acquires is extended to high-resolution video figure
Picture.
In two groups of video images, the picture frame of synchronization under same scene is found out, forms image pair；For each image
Right, the image block for carrying out 9 × 9 to two sub-pictures respectively operates；High-definition picture block forms high-resolution dictionary training sample
Collection, and low-resolution image block forms low-resolution dictionary training sample set.
9. according to the compressed sensing based infrared pedestrian image super-resolution rebuilding system of claim 8, which is characterized in that described
High-resolution dictionary and the training module of low-resolution dictionary are specifically used for:
High-resolution dictionary D is obtained by following formula traininghWith low-resolution dictionary Dl:
In formula, XhAnd YlIt is high-resolution sample set and low resolution sample set respectively；
During training dictionary, to progress while being trained using identical sparse coding algorithm come the image to acquisition, this
Process is described by following formula:
Wherein, N and M is respectively the dimension of high-resolution and low-resolution image block vector in sample database establishment process.
10. compressed sensing based infrared pedestrian image super resolution ratio reconstruction method according to claim 9, feature exist
In the rarefaction representation module is specifically used for:
By low-resolution image I1Interpolation amplification is to desired high-definition picture same size, then by amplified figure
Low-resolution image block is constituted as carrying out piecemeal；
The low-resolution image block of acquisition solve its it is low explanation dictionary characterization under rarefaction representation vector α, the rarefaction representation to
Amount is the corresponding rarefaction representation vector of corresponding high-definition picture block；
The reconstruction module is specifically used for:
The high-resolution dictionary D obtained by obtained image block rarefaction representation vector α and trainingh, reconstruct high-definition picture
Block, calculation formula are as follows: Dhα；
Calculated high-definition picture block, is spliced into complete image；And image is done to the boundary between image block
Smoothing processing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811644529.1A CN109785235A (en) | 2018-12-29 | 2018-12-29 | Compressed sensing based infrared pedestrian image super resolution ratio reconstruction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811644529.1A CN109785235A (en) | 2018-12-29 | 2018-12-29 | Compressed sensing based infrared pedestrian image super resolution ratio reconstruction method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109785235A true CN109785235A (en) | 2019-05-21 |
Family
ID=66499561
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811644529.1A Pending CN109785235A (en) | 2018-12-29 | 2018-12-29 | Compressed sensing based infrared pedestrian image super resolution ratio reconstruction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109785235A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112837220A (en) * | 2021-01-21 | 2021-05-25 | 华北电力大学（保定） | Method for improving resolution of infrared image and application thereof |
| CN113393374A (en) * | 2020-03-11 | 2021-09-14 | 联发科技股份有限公司 | Video processing circuit and method for performing SR operation |
| US20210287340A1 (en) * | 2020-03-11 | 2021-09-16 | Mediatek Inc. | Online ai super resolution engine and operations |
| CN113767415A (en) * | 2020-04-01 | 2021-12-07 | 京东方科技集团股份有限公司 | Computer-implemented method, apparatus, and computer program product |
| KR20220062575A (en) * | 2019-09-09 | 2022-05-17 | 엔비디아 코포레이션 | Video upsampling using one or more neural networks |
| CN117436937A (en) * | 2023-12-21 | 2024-01-23 | 暗物智能科技(广州)有限公司 | Path prediction method and system considering pedestrian portrait |
| CN117436937B (en) * | 2023-12-21 | 2024-07-05 | 暗物智能科技(广州)有限公司 | Path prediction method and system considering pedestrian portrait |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101971620A (en) * | 2008-02-29 | 2011-02-09 | 慧视科技有限公司 | A method of recording quality images |
| CN104680502A (en) * | 2015-03-19 | 2015-06-03 | 四川大学 | Infrared image super-resolution reconstruction method based on sparse dictionary and non-subsample Contourlet transform |
| CN104766273A (en) * | 2015-04-20 | 2015-07-08 | 重庆大学 | Infrared image super-resolution reestablishing method based on compressed sensing theory |
| CN104778671A (en) * | 2015-04-21 | 2015-07-15 | 重庆大学 | Image super-resolution method based on SAE and sparse representation |
| CN105678728A (en) * | 2016-01-19 | 2016-06-15 | 西安电子科技大学 | High-efficiency super-resolution imaging device and method with regional management |
| CN105844590A (en) * | 2016-03-23 | 2016-08-10 | 武汉理工大学 | Image super-resolution reconstruction method and system based on sparse representation |
| CN106934766A (en) * | 2017-03-15 | 2017-07-07 | 西安理工大学 | A kind of infrared image super resolution ratio reconstruction method based on rarefaction representation |
| CN107767337A (en) * | 2017-09-29 | 2018-03-06 | 深圳先进技术研究院 | A kind of image super-resolution rebuilding method, device, mobile terminal and storage medium |
-
2018
- 2018-12-29 CN CN201811644529.1A patent/CN109785235A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101971620A (en) * | 2008-02-29 | 2011-02-09 | 慧视科技有限公司 | A method of recording quality images |
| CN104680502A (en) * | 2015-03-19 | 2015-06-03 | 四川大学 | Infrared image super-resolution reconstruction method based on sparse dictionary and non-subsample Contourlet transform |
| CN104766273A (en) * | 2015-04-20 | 2015-07-08 | 重庆大学 | Infrared image super-resolution reestablishing method based on compressed sensing theory |
| CN104778671A (en) * | 2015-04-21 | 2015-07-15 | 重庆大学 | Image super-resolution method based on SAE and sparse representation |
| CN105678728A (en) * | 2016-01-19 | 2016-06-15 | 西安电子科技大学 | High-efficiency super-resolution imaging device and method with regional management |
| CN105844590A (en) * | 2016-03-23 | 2016-08-10 | 武汉理工大学 | Image super-resolution reconstruction method and system based on sparse representation |
| CN106934766A (en) * | 2017-03-15 | 2017-07-07 | 西安理工大学 | A kind of infrared image super resolution ratio reconstruction method based on rarefaction representation |
| CN107767337A (en) * | 2017-09-29 | 2018-03-06 | 深圳先进技术研究院 | A kind of image super-resolution rebuilding method, device, mobile terminal and storage medium |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20220062575A (en) * | 2019-09-09 | 2022-05-17 | 엔비디아 코포레이션 | Video upsampling using one or more neural networks |
| KR102661953B1 (en) * | 2019-09-09 | 2024-05-08 | 엔비디아 코포레이션 | Video upsampling using one or more neural networks |
| US12033301B2 (en) | 2019-09-09 | 2024-07-09 | Nvidia Corporation | Video upsampling using one or more neural networks |
| CN113393374A (en) * | 2020-03-11 | 2021-09-14 | 联发科技股份有限公司 | Video processing circuit and method for performing SR operation |
| US20210287340A1 (en) * | 2020-03-11 | 2021-09-16 | Mediatek Inc. | Online ai super resolution engine and operations |
| US11836893B2 (en) * | 2020-03-11 | 2023-12-05 | Mediatek Inc. | Online AI super resolution engine and operations |
| CN113767415A (en) * | 2020-04-01 | 2021-12-07 | 京东方科技集团股份有限公司 | Computer-implemented method, apparatus, and computer program product |
| CN112837220A (en) * | 2021-01-21 | 2021-05-25 | 华北电力大学（保定） | Method for improving resolution of infrared image and application thereof |
| CN112837220B (en) * | 2021-01-21 | 2023-07-25 | 华北电力大学（保定） | Method for improving resolution of infrared image and application thereof |
| CN117436937A (en) * | 2023-12-21 | 2024-01-23 | 暗物智能科技(广州)有限公司 | Path prediction method and system considering pedestrian portrait |
| CN117436937B (en) * | 2023-12-21 | 2024-07-05 | 暗物智能科技(广州)有限公司 | Path prediction method and system considering pedestrian portrait |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109785235A (en) | Compressed sensing based infrared pedestrian image super resolution ratio reconstruction method and system | |
| Jiang et al. | Matrix factorization for low-rank tensor completion using framelet prior | |
| He et al. | Learning based compressed sensing for SAR image super-resolution | |
| CN102708576B (en) | Method for reconstructing partitioned images by compressive sensing on the basis of structural dictionaries | |
| CN107657217A (en) | The fusion method of infrared and visible light video based on moving object detection | |
| CN104574336B (en) | Super-resolution image reconstruction system based on adaptive sub- mould dictionary selection | |
| CN110501072A (en) | A kind of reconstructing method of the fast illuminated spectrum imaging system based on the constraint of tensor low-rank | |
| Chen et al. | Single-image super-resolution using multihypothesis prediction | |
| Yanshan et al. | OGSRN: Optical-guided super-resolution network for SAR image | |
| Fang et al. | Arbitrary scale super resolution network for satellite imagery | |
| Wang et al. | Group shuffle and spectral-spatial fusion for hyperspectral image super-resolution | |
| CN114359041A (en) | Light field image space super-resolution reconstruction method | |
| Yu et al. | Color polarization image super-resolution reconstruction via a cross-branch supervised learning strategy | |
| Tang et al. | Feedback refined local-global network for super-resolution of hyperspectral imagery | |
| Yang et al. | Fast multisensor infrared image super-resolution scheme with multiple regression models | |
| Zhou et al. | Cloud removal for optical remote sensing imagery using distortion coding network combined with compound loss functions | |
| Moustafa et al. | Acceleration of super-resolution for multispectral images using self-example learning and sparse representation | |
| CN117011357A (en) | Human body depth estimation method and system based on 3D motion flow and normal map constraint | |
| CN102298768B (en) | High-resolution image reconstruction method based on sparse samples | |
| CN102708556A (en) | Single image super resolution method on basis of maintenance of reversed image | |
| CN116183226A (en) | Bearing test bed vibration displacement measurement and modal analysis algorithm based on phase | |
| Li et al. | Non-uniform Compressive Sensing Imaging Based on Image Saliency | |
| CN104112294B (en) | Intensity correlation imaging high speed three-dimensional reconfiguration system and method based on sparse constraint | |
| Kang et al. | Self-learning-based low-quality single image super-resolution | |
| Al-Hassan et al. | Enhancing face recognition at a distance using super resolution |
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 |