CN109448035A - Infrared image and visible light image registration method based on deep learning - Google Patents
Infrared image and visible light image registration method based on deep learning Download PDFInfo
- Publication number
- CN109448035A CN109448035A CN201811349860.0A CN201811349860A CN109448035A CN 109448035 A CN109448035 A CN 109448035A CN 201811349860 A CN201811349860 A CN 201811349860A CN 109448035 A CN109448035 A CN 109448035A
- Authority
- CN
- China
- Prior art keywords
- image
- infrared image
- network
- infrared
- images
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06T2207/10048—Infrared image
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention belongs to image procossings and technical field of computer vision, are related to a kind of infrared image and visible light image registration method based on deep learning, including are obtained generating image according to visible images；The displacement vector at control point and the local deformation parameter of space transformer are obtained according to infrared image and generation image；Geometric distortion device and resampler obtain deformation pattern according to infrared image；Deformation pattern is sent into decision networks with generation image and is differentiated, and using the difference of two images as loss function backpropagation to optimize network；If decision networks cannot tell authentic specimen and the pseudo- sample from generator, then it is assumed that network has reached optimal, and obtained deformation pattern is last registration result；First stage of the invention generates infrared image using a small amount of supervision message, and second stage is based on convolutional neural networks and does deformable image registration, to realize being registrated between visible images and infrared image.
Description
Technical field
The invention belongs to image procossings and technical field of computer vision, are related to a kind of infrared image based on deep learning
With visible light image registration method.
Background technique
Visible images and infrared image belong to two different forms, and infrared image is imaged according to the heat radiation of object,
It is influenced with abundant thermal radiation information, and not by light, background clutter, image-forming range；But infrared image can lose line
The details such as reason, structure.Visible images have color abundant and texture information, however, visible images are highly susceptible to shine
It is bright, the influence of factors such as block.The information that infrared image and visible images include has a complementarity, it is complementary in detection and
Identification, the scene analysis of pattern-recognition, remote sensing images, medical image, modern military forces use, the fields such as night surveillance, which have, widely answers
With.For the complementary information for making full use of visible images and infrared image, being registrated two kinds of images is most basic and most critical step
Suddenly.Image registration is that a kind of optimal transformation is found in transformation space, and two images is made to be able to achieve conversion by the transformation.
Currently, the method about image registration is broadly divided into three kinds: the first is the method based on region, and such methods are straight
Processing image intensity value is connect, by calculating the similitude of sliding window pair in two images, and considers most like position conduct
Corresponding relationship；For second the method based on feature, such methods be defined by determining notable feature similarity constraint with
Find the spatial alternation between feature.But since the usual resolution ratio of infrared spectrum is low and fuzzy, it is difficult to extract reliable close
Key point feature, the method based on feature generally can not work in infrared and visual picture registration task well.The third is
Method based on convolutional neural networks, in existing method, registration image is to being identical image spectrum or difference very little
Image spectrum, the direct training convolutional neural networks of the difference of the feature based on extraction.However, due to infrared image and visible light figure
Huge imaging difference as between, the training method are not suitable for infrared and visible images registration problems.More than being based on
Analysis, it is believed that can will be seen that light image is converted to visible images with confrontation network is generated, be used under same form
Convolutional neural networks learn character representation of the image pair about local displacement, to realize infrared matching with visible images
It is quasi-.
Summary of the invention
In order to realize it is infrared with visible images be registrated the present invention propose a kind of infrared image based on deep learning with
Visible light image registration method, comprising the following steps:
S1, the infrared image of Same Scene and visible images are cropped to fixed dimension and are stitched together, will spliced
Image afterwards is divided into training set and test set；
S2, it will be seen that light image as input, corresponding infrared image as label training generation fight network, obtain by
The generation image that visible images generate；
S3, it is slided on infrared image and generation image using fixed dimension sliding window, by two images same position
Image block be sent into convolutional neural networks in pairs, obtain the displacement vector at control point and the local deformation parameter of space transformer；
S4, space transformer generate intensive displacement vector field according to local deformation parameter and the displacement vector at control point, so
Former infrared image is deformed into the similar deformation pattern for generating image using intensive displacement vector field by resampler afterwards；
S5, will differentiate in deformation pattern and generation image feeding decision networks, and using the difference of two images as loss letter
Number backpropagation is to optimize network；If decision networks cannot tell authentic specimen and the pseudo- sample from generator, recognize
Had reached for network optimal, obtained deformation pattern is last registration result.
Preferably, step S1 is specifically included: it will be seen that light image and infrared image are cropped to 256 × 256 fixed dimension,
Visible images pixel value is converted into [- 1,1] from [0,1]；The input value of visible images Color Channel turns from [- 110,110]
Turn to [- 1,1].
Preferably, it includes generating network and decision networks that network is fought in step S2, in generating network, it will be seen that light figure
Picture and random noise generate network according to multiple and different infrared of visible images and random noise as the input for generating network
Image puppet sample, i.e. generation image, wherein the data distribution for generating image obeys visible images data distribution；Decision networks pair
Image sources do two discriminant classifications: authentic specimen, i.e. infrared image are derived from, from the pseudo- sample of generator, i.e. generation figure
Picture；Decision networks differentiate the source for generating image and infrared image；If decision networks, which can be told, generates image and red
The difference for generating image and infrared image is fed back to generation network by the source of outer image, generates network, sentences
Certainly network and the feedback generated between network and decision networks constitute confrontation network.
Preferably, step S3 is specifically included:
S31, former infrared image is taken with sliding window and generates the image block of image same position, be sent into convolutional Neural in pairs
Network；It is 1 that sliding window, which takes step-length, traverses whole image；
S32, by the down-sampling layer of four alternate 3 × 3 zero padding convolutional layers and 2 × 2 average pond, retain defeated
Enter the high-order feature of image；
S33, learn to indicate local displacement important feature in image by 31 × 1 full articulamentums, and defeated
The deformation parameter of the corresponding infrared image in space and generation image pair out.
Preferably, the difference of two images is optimized network as loss function backpropagation includes: calculating deformation pattern
With the structural similarity SSIM between generation imageX,F, and by deformation pattern and generate the dissimilar type 1-SSIM between figureX,FMake
It is loss function backpropagation to optimize network；Structural similarity SSIM between two imagesX,FIs defined as:
Wherein, SSIMX,FIt indicates to generate the structural similarity between image X and deformation pattern F, x indicates that in size be M × N
Local window in generate image image block, f indicate size be M × N local window in deformation pattern image block；σx
Indicate the standard deviation of generation image, σxIndicate the standard deviation of deformation pattern, σxfIt is the standard association side for generating image and deformation pattern
Difference, μxIndicate the average value of generation image, μfIndicate the average value of deformation pattern；C1、C2And C3For algorithmic stability parameter.
Preferably, before carrying out the splicing of infrared image and visible images, gaussian filtering is carried out to visible images
Processing carries out linear enhancing processing to infrared image.
Innovation of the invention obscures compared with being infrared and infrared image huge with visible images morphological differences, this
Invention is infrared image visible light transformation by GAN network, to be compared under a kind of form.
Detailed description of the invention
Fig. 1 is that the present invention is based on the infrared images of deep learning and visible light image registration method overall flow schematic diagram；
Fig. 2 is that the present invention is based on the infrared images of deep learning and visible light image registration method integral frame schematic diagram；
Fig. 3 is convolutional neural networks structural schematic diagram of the invention；
Fig. 4 is the image of an infrared image and visible images splicing in the embodiment of the present invention；
Fig. 5 is the last registration result figure of Fig. 4 image in the embodiment of the present invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
The present invention provides a kind of infrared image and visible light image registration method based on deep learning, such as Fig. 1, including with
Lower step:
S1, the infrared image of Same Scene and visible images are cropped to fixed dimension and are stitched together, will spliced
Image afterwards is divided into training set and test set；
S2, using the visible images of image as input, the infrared image of image fights network as label training generation,
Obtain the generation image of the visible images；
S3, it is slided on infrared image and generation image using fixed dimension sliding window, by two images same position
Image block be sent into convolutional neural networks in pairs, obtain the displacement vector at control point and the local deformation parameter of space transformer；
S4, space transformer generate intensive displacement vector field according to local deformation parameter and the displacement vector at control point, so
Infrared image is deformed into the similar deformation pattern for generating image using intensive displacement vector field by resampler afterwards；
S5, will differentiate in deformation pattern and generation image feeding decision networks, and using the difference of two images as loss letter
Number backpropagation is to optimize convolutional neural networks；If decision networks cannot tell authentic specimen and the pseudo- sample from generator
This, then it is assumed that network has reached optimal, and obtained deformation pattern is last registration result.
The present invention provides a kind of specific embodiment, the data packet used in the present embodiment include under 12 scenes can
Light-exposed and infrared image shares 400 pairs of authentic specimens, it will be seen that light image is cropped to 256 × 256 fixed ruler with infrared image
It is very little, it is seen that light image pixel value is converted into [- 1,1] from [0,1]；The input value of visible images Color Channel is from [- 110,110]
It is converted into [- 1,1]；It is the conversion side of visible images that wherein visible images, which are authentic specimen infrared image as training label,
Be to, it is seen that light image it is colored, to be transformed into infrared image color gamut and need to compress, and be by the expansion of the range of pixel value
In order to which contrast is obvious.
It will be seen that light image, as input, corresponding infrared image fights network as label training generation, obtain by can
The generation image that light-exposed image generates；Visible images and random noise generate network and generate not as the input for generating network
Same pseudo- sample, data distribution all obey visible images data distribution as far as possible, and decision networks do two classification to image sources
Differentiate: from authentic specimen, from the pseudo- sample of generator.Generator and decision device are confronted with each other and iteration optimization, most
Obtain eventually decision device can not correct decision source pseudo- sample, that is, the infrared image generated.If decision networks can be told really
The difference of authentic specimen and pseudo- sample is fed back to generation network as loss function, generates net by the source of sample and pseudo- sample
Network, decision networks and the feedback generated between network and decision networks constitute confrontation network.
It is slided on infrared image and generation image using fixed dimension sliding window, by the figure of two images same position
As block is sent into convolutional neural networks in pairs, the displacement vector at control point and the local deformation parameter of space transformer are obtained；Convolution
Neural network is as shown in figure 3, include the 1st layer~the 7th layer to be linked in sequence, wherein the 1st layer~the 4th layer by convolutional layer and pond
Layer is constituted, and the 5th layer~the 7th layer is made of full articulamentum；The specific generating process of local deformation parameter includes:
S31, former infrared image is taken with sliding window and generates the image block of image same position, be sent into convolutional Neural in pairs
Network；It is 1 that sliding window, which takes step-length, traverses whole image；
S32, by the down-sampling layer of four alternate 3 × 3 zero padding convolutional layers and 2 × 2 average pond, retain defeated
Enter the high-order feature of image, that is, the 1st layer~the 4th layer for passing through convolutional neural networks；
S33, by 31 × 1 full articulamentums, that is, pass through the 5th layer~the 7th layer of convolutional neural networks, learn in image
Local displacement important feature is indicated, and exports the corresponding infrared image in space and generates the deformation parameter of image pair.
Space transformer generates intensive displacement vector field according to local deformation parameter and the displacement vector at control point, then weighs
Former infrared image is deformed into the similar deformation pattern for generating image using intensive displacement vector field by sampler；By convolutional Neural net
The deformation parameter of network output and the displacement vector at control point, space transformer generate intensive displacement vector field, resampler benefit
Former infrared image is deformed into the similar deformation pattern for generating image with displacement vector field.
Deformation pattern is sent into decision networks with generation image to be differentiated, and the difference of two images is reversed as loss function
It propagates to optimize network；If decision networks cannot tell authentic specimen and the pseudo- sample from generator, then it is assumed that network
Have reached optimal, obtained deformation pattern is last registration result.
It calculates deformation pattern and generates the similitude between image, optimize network by backpropagation dissimilarity, tie
The measurement of structure index of similarity is defined as:
Structural similarity index measurement mainly consists of three parts: impairment correlations, luminance distortion and contrast distortion；Its
In, SSIMX,FIt indicates to generate the structural similarity between image X and deformation pattern F, x is indicated in the local window that size is M × N
The middle image block for generating image, f indicate the image block of the deformation pattern in the local window that size is M × N；σxIt indicates to generate figure
The standard deviation of picture, σfIndicate the standard deviation of deformation pattern, σxfIt is the standard covariance for generating image and deformation pattern, μxIndicate life
At the average value of image, μfIndicate the average value of deformation pattern；C1、C2And C3For algorithmic stability parameter.
Fig. 4 is the image that visible images and infrared image are spliced, and Fig. 5 is after the image realization in Fig. 4 is registrated
Deformation pattern；It can be seen from the figure that this method can be preferable to realize in the case where complex background and low contrast
Visible images and infrared image registration.
It although an embodiment of the present invention has been shown and described, for the ordinary skill in the art, can be with
A variety of variations, modification, replacement can be carried out to these embodiments without departing from the principles and spirit of the present invention by understanding
And modification, the scope of the present invention is defined by the appended.
Claims (7)
1. infrared image and visible light image registration method based on deep learning, which comprises the following steps:
S1, the infrared image of Same Scene and visible images are cropped to fixed dimension and are stitched together, it will be spliced
Image is divided into training set and test set；
S2, using the visible images of image as input, the infrared image of image fights network as label training generation, right
The generation image of the visible images is obtained in anti-network；
S3, it is slided on infrared image and generation image using fixed dimension sliding window, by the figure of two images same position
As block is sent into convolutional neural networks in pairs, the displacement vector at control point and the local deformation parameter of space transformer are obtained；
S4, space transformer generate intensive displacement vector field according to local deformation parameter and the displacement vector at control point, then weigh
Infrared image is deformed into the similar deformation pattern for generating image using intensive displacement vector field by sampler；
S5, will differentiate in deformation pattern and generation image feeding decision networks, and the difference of two images is anti-as loss function
To propagation to optimize convolutional neural networks；If decision networks cannot tell authentic specimen and the pseudo- sample from generator,
It is optimal then to think that network has reached, obtained deformation pattern is last registration result.
2. the infrared image and visible light image registration method according to claim 1 based on deep learning, feature exist
In infrared image and visible images are cropped to 256 × 256 fixed dimension in step S1.
3. the infrared image and visible light image registration method according to claim 2 based on deep learning, feature exist
In the visible light picture compression after reduction being compressed in the color gamut of infrared image, i.e., by visible images Color Channel
Input value be converted into [- 1,1] from [- 110,110], and will be seen that light image pixel value [- 1,1] is converted into from [0,1].
4. the infrared image and visible light image registration method according to claim 1 based on deep learning, feature exist
In fighting network in step S2 includes generating network and decision networks, in generating network, it will be seen that light image and random noise
As the input for generating network, network is generated according to the pseudo- sample of the multiple and different infrared image of visible images and random noise
This, that is, generate image, wherein the data distribution for generating image obeys visible images data distribution；Decision networks are to image sources
Do two discriminant classifications: from authentic specimen or from the pseudo- sample of generator；In the training stage, decision networks scheme generation
The source of picture and infrared image is differentiated；If decision networks can tell the source for generating image and infrared image, will generate
The difference of image and infrared image feeds back to generation network as loss function, generate network, decision networks and generate network with
Feedback between decision networks collectively forms confrontation network.
5. the infrared image and visible light image registration method according to claim 1 based on deep learning, feature exist
In step S3 is specifically included:
S31, former infrared image is taken with sliding window and generates the image block of image same position, be sent into convolutional Neural net in pairs
Network；It is 1 that sliding window, which takes step-length, traverses whole image；
S32, by the down-sampling layer of four alternate 3 × 3 zero padding convolutional layers and 2 × 2 average pond, retain input figure
The high-order feature of picture；
S33, learn to indicate local displacement important feature in image by 31 × 1 full articulamentums, and export sky
Between corresponding former infrared image and generate the deformation parameter of image pair.
6. the infrared image and visible light image registration method according to claim 1 based on deep learning, feature exist
In described the difference of two images is optimized network as loss function backpropagation includes: to calculate deformation pattern and generation figure
Structural similarity SSIM as betweenX,F, and by deformation pattern and generate the dissimilar type 1-SSIM between figureX,FAs loss letter
Number backpropagation is to optimize network；Structural similarity SSIM between two imagesX,FIs defined as:
Wherein, SSIMX,FIt indicates to generate the structural similarity between image X and deformation pattern F, x is indicated in the office that size is M × N
The image block of image is generated in portion's window, f indicates the image block of the deformation pattern in the local window that size is M × N；σxIt indicates
Generate the standard deviation of image, σxIndicate the standard deviation of deformation pattern, σxfIt is the standard covariance for generating image and deformation pattern, μx
Indicate the average value of generation image, μfIndicate the average value of deformation pattern；C1、C2And C3For algorithmic stability parameter.
7. the infrared image and visible light image registration method according to claim 1 based on deep learning, feature exist
In before carrying out the splicing of infrared image and visible images, to visible images progress gaussian filtering process, to infrared figure
As carrying out linear enhancing processing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811349860.0A CN109448035A (en) | 2018-11-14 | 2018-11-14 | Infrared image and visible light image registration method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811349860.0A CN109448035A (en) | 2018-11-14 | 2018-11-14 | Infrared image and visible light image registration method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109448035A true CN109448035A (en) | 2019-03-08 |
Family
ID=65552283
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811349860.0A Pending CN109448035A (en) | 2018-11-14 | 2018-11-14 | Infrared image and visible light image registration method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109448035A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109903299A (en) * | 2019-04-02 | 2019-06-18 | 中国矿业大学 | A kind of conditional generates the heterologous remote sensing image registration method and device of confrontation network |
| CN110021037A (en) * | 2019-04-17 | 2019-07-16 | 南昌航空大学 | A kind of image non-rigid registration method and system based on generation confrontation network |
| CN110472089A (en) * | 2019-08-16 | 2019-11-19 | 重庆邮电大学 | A kind of infrared and visible images search method generating network based on confrontation |
| CN111193920A (en) * | 2019-12-31 | 2020-05-22 | 重庆特斯联智慧科技股份有限公司 | Video picture three-dimensional splicing method and system based on deep learning network |
| CN112418163A (en) * | 2020-12-09 | 2021-02-26 | 北京深睿博联科技有限责任公司 | Multispectral target detection blind guiding system |
| CN112487233A (en) * | 2020-11-27 | 2021-03-12 | 重庆邮电大学 | Infrared and visible light image retrieval method based on feature decoupling |
| CN113379658A (en) * | 2021-06-01 | 2021-09-10 | 大连海事大学 | Unmanned aerial vehicle observation target feature double-light fusion method and system |
| CN114820733A (en) * | 2022-04-21 | 2022-07-29 | 北京航空航天大学 | Interpretable thermal infrared visible light image registration method and system |
| CN114881884A (en) * | 2022-05-24 | 2022-08-09 | 河南科技大学 | Infrared target sample enhancement method based on generation countermeasure network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107154014A (en) * | 2017-04-27 | 2017-09-12 | 上海大学 | A kind of real-time color and depth Panorama Mosaic method |
| CN108230375A (en) * | 2017-12-27 | 2018-06-29 | 南京理工大学 | Visible images and SAR image registration method based on structural similarity fast robust |
| CN108510532A (en) * | 2018-03-30 | 2018-09-07 | 西安电子科技大学 | Optics and SAR image registration method based on depth convolution GAN |
-
2018
- 2018-11-14 CN CN201811349860.0A patent/CN109448035A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107154014A (en) * | 2017-04-27 | 2017-09-12 | 上海大学 | A kind of real-time color and depth Panorama Mosaic method |
| CN108230375A (en) * | 2017-12-27 | 2018-06-29 | 南京理工大学 | Visible images and SAR image registration method based on structural similarity fast robust |
| CN108510532A (en) * | 2018-03-30 | 2018-09-07 | 西安电子科技大学 | Optics and SAR image registration method based on depth convolution GAN |
Non-Patent Citations (2)
| Title |
|---|
| BOB D.DE VOS ETC.: ""End-to-End Unsupervised Deformable Image Registration with a Convolutional Neural Network"", 《DEEP LEARNING IN MEDICAL IMAGE ANALYSIS AND MULTIMODAL LEARNING FOR CLINICAL DECISION SUPPORT》 * |
| LAN WANG ETC.: ""Infrared and Visible Image Registration Using Transformer Adversarial Network"", 《2018 25TH IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109903299A (en) * | 2019-04-02 | 2019-06-18 | 中国矿业大学 | A kind of conditional generates the heterologous remote sensing image registration method and device of confrontation network |
| CN109903299B (en) * | 2019-04-02 | 2021-01-05 | 中国矿业大学 | Registration method and device for heterogenous remote sensing image of conditional generation countermeasure network |
| CN110021037A (en) * | 2019-04-17 | 2019-07-16 | 南昌航空大学 | A kind of image non-rigid registration method and system based on generation confrontation network |
| CN110472089A (en) * | 2019-08-16 | 2019-11-19 | 重庆邮电大学 | A kind of infrared and visible images search method generating network based on confrontation |
| CN110472089B (en) * | 2019-08-16 | 2022-03-11 | 重庆邮电大学 | Infrared and visible light image retrieval method based on countermeasure generation network |
| CN111193920A (en) * | 2019-12-31 | 2020-05-22 | 重庆特斯联智慧科技股份有限公司 | Video picture three-dimensional splicing method and system based on deep learning network |
| CN112487233A (en) * | 2020-11-27 | 2021-03-12 | 重庆邮电大学 | Infrared and visible light image retrieval method based on feature decoupling |
| CN112487233B (en) * | 2020-11-27 | 2022-07-12 | 重庆邮电大学 | Infrared and visible light image retrieval method based on feature decoupling |
| CN112418163A (en) * | 2020-12-09 | 2021-02-26 | 北京深睿博联科技有限责任公司 | Multispectral target detection blind guiding system |
| CN113379658A (en) * | 2021-06-01 | 2021-09-10 | 大连海事大学 | Unmanned aerial vehicle observation target feature double-light fusion method and system |
| CN113379658B (en) * | 2021-06-01 | 2024-03-15 | 大连海事大学 | Unmanned aerial vehicle observation target feature double-light fusion method and system |
| CN114820733A (en) * | 2022-04-21 | 2022-07-29 | 北京航空航天大学 | Interpretable thermal infrared visible light image registration method and system |
| CN114820733B (en) * | 2022-04-21 | 2024-05-31 | 北京航空航天大学 | Interpretable thermal infrared visible light image registration method and system |
| CN114881884A (en) * | 2022-05-24 | 2022-08-09 | 河南科技大学 | Infrared target sample enhancement method based on generation countermeasure network |
| CN114881884B (en) * | 2022-05-24 | 2024-03-29 | 河南科技大学 | Infrared target sample enhancement method based on generation countermeasure network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109448035A (en) | Infrared image and visible light image registration method based on deep learning | |
| CN108537743B (en) | Face image enhancement method based on generation countermeasure network | |
| CN110991281B (en) | Dynamic face recognition method | |
| CN112733950A (en) | Power equipment fault diagnosis method based on combination of image fusion and target detection | |
| CN110135375A (en) | More people's Attitude estimation methods based on global information integration | |
| CN107463920A (en) | A kind of face identification method for eliminating partial occlusion thing and influenceing | |
| CN108921051A (en) | Pedestrian's Attribute Recognition network and technology based on Recognition with Recurrent Neural Network attention model | |
| CN110647874B (en) | End-to-end blood cell identification model construction method and application | |
| CN111145131A (en) | Infrared and visible light image fusion method based on multi-scale generation type countermeasure network | |
| CN110097528A (en) | A kind of image interfusion method based on joint convolution autoencoder network | |
| CN112784736B (en) | Character interaction behavior recognition method based on multi-modal feature fusion | |
| CN106303233A (en) | A kind of video method for secret protection merged based on expression | |
| CN107609464B (en) | A kind of real-time face rapid detection method | |
| CN106155299A (en) | A kind of method and device that smart machine is carried out gesture control | |
| CN110390308A (en) | It is a kind of to fight the video behavior recognition methods for generating network based on space-time | |
| CN109741285A (en) | A kind of construction method and system of underwater picture data set | |
| Xing et al. | Multi-level adaptive perception guidance based infrared and visible image fusion | |
| CN105469413B (en) | It is a kind of based on normalization ring weighting without refer to smear restoration image synthesis method for evaluating quality | |
| Guo et al. | An efficient dual spatial–spectral fusion network | |
| CN114663795A (en) | Target detection method for obtaining rear image of glass curtain wall by range gating imaging equipment | |
| CN106127754A (en) | CME detection method based on fusion feature and space-time expending decision rule | |
| CN111178229B (en) | Deep learning-based vein imaging method and device | |
| CN109064444A (en) | Track plates Defect inspection method based on significance analysis | |
| Tang et al. | MPCFusion: Multi-scale parallel cross fusion for infrared and visible images via convolution and vision Transformer | |
| CN103020995B (en) | A kind of object feature comparison method being applicable to search for target in video |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190308 |
|
| RJ01 | Rejection of invention patent application after publication |