CN109544492A - A kind of multi-focus image fusion data set production method based on convolutional neural networks - Google Patents
A kind of multi-focus image fusion data set production method based on convolutional neural networks Download PDFInfo
- Publication number
- CN109544492A CN109544492A CN201811254081.2A CN201811254081A CN109544492A CN 109544492 A CN109544492 A CN 109544492A CN 201811254081 A CN201811254081 A CN 201811254081A CN 109544492 A CN109544492 A CN 109544492A
- Authority
- CN
- China
- Prior art keywords
- image
- focusing
- data set
- convolutional neural
- neural networks
- 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
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 23
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 21
- 230000004927 fusion Effects 0.000 title claims abstract description 18
- 238000012549 training Methods 0.000 claims abstract description 8
- 238000000034 method Methods 0.000 claims description 6
- 230000000694 effects Effects 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 3
- 238000012360 testing method Methods 0.000 abstract description 3
- 230000011218 segmentation Effects 0.000 abstract description 2
- 238000013528 artificial neural network Methods 0.000 abstract 1
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
-
- 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
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The multi-focus image fusion data set production method based on convolutional neural networks that the invention discloses a kind of, first building data set, the size and format of unified image；And Fuzzy Processing is carried out to focusedimage and generates non-focusing blurred picture；Then one group of focusing and non-focusing image are split and are spliced, finally make image tag to be focused the segmentation with non-focusing region to meet the requirement needed based on the image for dividing network while comprising focused pixel and non-focusing pixel.Can effectively solve the problem that presently, there are study the data set size and format disunity that no workable standard data set and some researchers provide for multi-focus image fusion, the problems such as quantity can not meet training neural network based and test request.
Description
Technical field
The invention belongs to image procossings and image data manufacture technology field, and in particular to one kind is suitable for based on convolution mind
The production method of multi-focus image fusion data set through network area segmentation.
Background technique
Multi-focus image fusion is more mainly for the treatment of obtaining from same target scene as a kind of informix means
Width focuses on the source images of different zones.Image information utilization rate can be effectively improved with multi-focus image fusion, is guaranteed
The reliability and accuracy of system subsequent processing, thus it is widely used in the fields such as computer vision, medical imaging, military affairs.
The features such as Pixel-level multi-focus image fusion is due to raw information and high minutia reservation degree and be considered as multiple focussing image and melt
The basis of conjunction.Focused pixel and non-focusing pixel in correct classification source images are the key that Pixel-level multi-focus image fusions.
In recent years, with the promotion of hardware computing capability, theoretical development and the arriving of big data, convolutional Neural net
Network has obtained extensive research, and the model of some convolutional neural networks is used for the training identification of image, voice, and achieves very well
Effect.Due to the introducing of GPU, model very complicated in the past passes through parallel computation now to be easily trained to, and contract significantly
In the short trained period for adjusting ginseng, improve the practicability of convolutional neural networks.
Currently, some scholars also attempt to merge multiple focussing image using convolutional neural networks, but convolutional Neural net
Network needs huge training set and test set to complete the training of model, not can be used for convolutional neural networks training largely but
Non-focusing image data set.
Summary of the invention
In order to overcome the deficiencies in the prior art, the present invention proposes a kind of multi-focus figure based on convolutional neural networks
As fused data set production method.Pass through Gaussian Blur according to the clear image of some focusing, then is used different methods
It carries out the processing such as splicing, then obtains that the multiple focussing image that classical convolutional neural networks are trained, test can be directly placed into
Data set.
A kind of multi-focus image fusion data set production method based on convolutional neural networks, includes the following steps:
S1: building focusedimage data set, and unification is carried out to picture specification；
S2: fuzzy operation is carried out to all images in focusedimage data set one by one, non-focusing image is obtained, will obscure
The corresponding focusedimage in operation front and back and non-focusing image are as a pair of of data；
S3: it to the focusedimage and non-focusing image in each pair of data, is obtained respectively with the symmetry axis cutting of its two-dimensional surface
Half figure pairs of to two groups, and the focusing of same area half is schemed to scheme with non-focusing half as one and Ban figures；
S4: taking half figure of non-focusing in one group using symmetry axis is axis overturning again with same group of half figure of focusing along symmetry axis split,
It obtains focusing a stitching image with non-focusing effect comprising same area simultaneously；
S5: the binaryzation label figure of production size identical as the stitching image in step S4.
Further, the image data set in the step S1 further includes that a small amount of scholar studies disclosed image data set,
The disclosure image data is concentrated comprising homologous focusedimage and non-focusing image as a pair of of data.
Further, the public image data set includes Multifocus Image Dataset, Lytro
The public image collection that Dataset and Multifocus Image Dataset is provided.
Further, picture size is reduced to 260 × 260 unification by down-sampling by the picture specification in the step S1
The RGB image of specification.
Further, the progress fuzzy operation in the step S2 refers to is filtered using the border circular areas mean value that MATLAB is provided
Wave device does smothing filtering to focusedimage, to obtain non-focusing image.
Further, the symmetry axis for carrying out cutting respectively to focusedimage and non-focusing image in the step S3 includes to hang down
Histogram is to, horizontal direction and diagonal.
Further, the manufacturing process of the binaryzation label figure of the step S5 specifically: define in one and step S4
Stitching image equidimension binary map, which corresponds to the focal zone value 1 of stitching image as white, corresponding spliced map
It is black that the non-focusing image-region value of picture, which is 0, obtains instruction convolutional neural networks training image and focuses and non-focusing region
Label figure.
Compared with prior art, the beneficial effects of the present invention are: solve the poly currently based on convolutional neural networks
Burnt image co-registration data volume is insufficient, can not be put into the problems such as convolutional neural networks training.
Detailed description of the invention
Fig. 1 is flow diagram of the invention；
Fig. 2 is the production schematic diagram of one group of sample cutting, combination in the present invention.
Specific embodiment
Xia Mianjiehefutuhejuti compare Jia Shishifangshiduibenfamingzuojinyibuxiangxishuoming.
As shown in Figure 1, the present invention provides a kind of multi-focus image fusion data set production side based on convolutional neural networks
Method includes the following steps:
Step 1: building focusedimage data set, and unification is carried out to picture specification；
Step 2: by size in focusedimage data set different gray level image and RGB image, by down-sampling by image
Long size is reduced to 260 × 260 square image.
Step 3: fuzzy operation being carried out to focusedimage using the common image library of image procossing, to reach non-focusing effect
Fruit.The border circular areas mean filter provided using MATLAB software does smothing filtering, Filtering Template configuration ginseng to focusedimage
Number r ∈ { 2+i/8；I=1,2 ... 8 } and to each image random value, the value is related with border circular areas radius.It is non-that treated
Focusedimage and focusedimage are as a pair of of data.
Step 4: a pair of focus is obtained into two groups of half pairs of figures with non-focusing image with the symmetry axis cutting of its two-dimensional surface,
The non-focusing of same area half is schemed and is focused half figure and is used as one group, and cutting detailed process is as shown in Fig. 2 201,202,203.
Wherein, 201, a pair of of image of the homologous acquisition of selection, wherein a width is focusedimage A, another width is non-focusing image
B。
202, two images are subjected to cutting according to symmetry axis.It, can be respectively according to vertical since image is square
Directly, four symmetry axis of horizontal and diagonal carry out cutting.Pair of homologous image A and B in this way is obtained with 8 groups and includes same zone
Domain non-focusing half is schemed and is focused half and schemes.
203, the focusing of same area half is schemed and non-focusing half is schemed as one and Ban figures.
Step 5: half figure of non-focusing in one group being taken to spell again with same group of half figure of focusing along symmetry axis using symmetry axis as axis overturning
It closes, so that a stitching image simultaneously comprising same area focusing and non-focusing effect is obtained, it is specific as 204,205 in Fig. 2
It is shown；
Step 6: production binaryzation label figure defines the binary map of one with stitching image equidimension in step 4, corresponding to spell
Image focus area value 1 is connect as white, it is black that non-focusing region value, which is 0, thus obtains instruction convolutional neural networks instruction
Practice the label figure that image focuses (white) and non-focusing region (black).
For some researchers, including Multifocus Image Dataset and Lytro are derived from step 1 on a small quantity
Some public image collection that Dataset, Multifocus Image Dataset are provided, the figure in data set provided due to it
As including homologous focusedimage and non-focusing image, it is therefore desirable to obtain every group of image respectively by the method for manual markings
Focusing and non-focusing region indicating label figure, for non-focusing image therein, do not need to carry out fuzzy operation obtain it is non-poly-
Burnt image, it is only necessary to find its homologous focusedimage and non-focusing image, by the format of focusedimage and non-focusing image into
Row is unified and as a pair of of data, and operation is continued in step 4 according to the invention, 5,6.
The preferred embodiment of the present invention has been described above in detail, still, during present invention is not limited to the embodiments described above
Detail a variety of equivalents can be carried out to technical solution of the present invention within the scope of the technical concept of the present invention, this
A little equivalents all belong to the scope of protection of the present invention.
Claims (7)
1. a kind of multi-focus image fusion data set production method based on convolutional neural networks, it is characterised in that: including as follows
Step:
S1: building focusedimage data set, and unification is carried out to picture specification；
S2: fuzzy operation is carried out to all images in focusedimage data set one by one, non-focusing image is obtained, by fuzzy operation
The corresponding focusedimage in front and back and non-focusing image are as a pair of of data；
S3: to the focusedimage and non-focusing image in each pair of data, two are obtained with the symmetry axis cutting of its two-dimensional surface respectively
Half pairs of figure of group, and the focusing of same area half is schemed to scheme with non-focusing half as one and Ban figures；
S4: it takes half figure of non-focusing in one group to overturn by axis of symmetry axis again with same group of half figure of focusing along symmetry axis split, obtains
A stitching image with non-focusing effect is focused comprising same area simultaneously；
S5: the binaryzation label figure of production size identical as the stitching image in step S4.
2. a kind of multi-focus image fusion data set production method based on convolutional neural networks according to claim 1,
It is characterized in that the image data set in the step S1 further includes that this field scholar studies disclosed image data set, the disclosure
Image data is concentrated comprising homologous focusedimage and non-focusing image as a pair of of data.
3. a kind of multi-focus image fusion data set production side based on convolutional neural networks according to claim 2
Method, it is characterised in that the public image data set include Multifocus Image Dataset, Lytro Dataset and
The public image collection that Multifocus Image Dataset is provided.
4. a kind of multi-focus image fusion data set production method based on convolutional neural networks according to claim 1,
It is characterized by: picture size is reduced to 260 × 260 unified specification by down-sampling by picture specification in the step S1
RGB image.
5. a kind of multi-focus image fusion data set production method based on convolutional neural networks according to claim 1,
It is characterized by: the progress fuzzy operation in the step S2 refers to the border circular areas mean filter pair provided using MATLAB
Focusedimage does smothing filtering, to obtain non-focusing image.
6. a kind of multi-focus image fusion data set production method based on convolutional neural networks according to claim 1,
It is characterized by: the symmetry axis for carrying out cutting respectively to focusedimage and non-focusing image in the step S3 includes Vertical Square
To, horizontal direction and diagonal.
7. a kind of multi-focus image fusion data set production method based on convolutional neural networks according to claim 1,
It is characterized by: the manufacturing process of the binaryzation label figure of the step S5 specifically: define one and the splicing in step S4
The binary map of image equidimension, the binary map correspond to the focal zone value 1 of stitching image as white, correspond to the non-of stitching image
It is black that focusedimage region value, which is 0, obtains the label of instruction convolutional neural networks training image focusing and non-focusing region
Figure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811254081.2A CN109544492A (en) | 2018-10-25 | 2018-10-25 | A kind of multi-focus image fusion data set production method based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811254081.2A CN109544492A (en) | 2018-10-25 | 2018-10-25 | A kind of multi-focus image fusion data set production method based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109544492A true CN109544492A (en) | 2019-03-29 |
Family
ID=65845205
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811254081.2A Pending CN109544492A (en) | 2018-10-25 | 2018-10-25 | A kind of multi-focus image fusion data set production method based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109544492A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110031477A (en) * | 2019-04-04 | 2019-07-19 | 中设设计集团股份有限公司 | Bridge key component disease early warning system and method based on image monitoring data |
| CN112750097A (en) * | 2021-01-14 | 2021-05-04 | 中北大学 | Multi-modal medical image fusion based on multi-CNN combination and fuzzy neural network |
| CN112926354A (en) * | 2019-12-05 | 2021-06-08 | 北京超星未来科技有限公司 | Deep learning-based lane line detection method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897987A (en) * | 2017-01-18 | 2017-06-27 | 江南大学 | Image interfusion method based on translation invariant shearing wave and stack own coding |
| CN107103331A (en) * | 2017-04-01 | 2017-08-29 | 中北大学 | A kind of image interfusion method based on deep learning |
| CN107273896A (en) * | 2017-06-15 | 2017-10-20 | 浙江南自智能科技股份有限公司 | A kind of car plate detection recognition methods based on image recognition |
| CN108537746A (en) * | 2018-03-21 | 2018-09-14 | 华南理工大学 | A kind of fuzzy variable method for blindly restoring image based on depth convolutional network |
| CN108596222A (en) * | 2018-04-11 | 2018-09-28 | 西安电子科技大学 | Image interfusion method based on deconvolution neural network |
-
2018
- 2018-10-25 CN CN201811254081.2A patent/CN109544492A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897987A (en) * | 2017-01-18 | 2017-06-27 | 江南大学 | Image interfusion method based on translation invariant shearing wave and stack own coding |
| CN107103331A (en) * | 2017-04-01 | 2017-08-29 | 中北大学 | A kind of image interfusion method based on deep learning |
| CN107273896A (en) * | 2017-06-15 | 2017-10-20 | 浙江南自智能科技股份有限公司 | A kind of car plate detection recognition methods based on image recognition |
| CN108537746A (en) * | 2018-03-21 | 2018-09-14 | 华南理工大学 | A kind of fuzzy variable method for blindly restoring image based on depth convolutional network |
| CN108596222A (en) * | 2018-04-11 | 2018-09-28 | 西安电子科技大学 | Image interfusion method based on deconvolution neural network |
Non-Patent Citations (1)
| Title |
|---|
| 李子恒: "像素级多聚焦图像融合算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110031477A (en) * | 2019-04-04 | 2019-07-19 | 中设设计集团股份有限公司 | Bridge key component disease early warning system and method based on image monitoring data |
| CN112926354A (en) * | 2019-12-05 | 2021-06-08 | 北京超星未来科技有限公司 | Deep learning-based lane line detection method and device |
| CN112750097A (en) * | 2021-01-14 | 2021-05-04 | 中北大学 | Multi-modal medical image fusion based on multi-CNN combination and fuzzy neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TWI777092B (en) | Image processing method, electronic device, and storage medium | |
| Jaw et al. | DesnowGAN: An efficient single image snow removal framework using cross-resolution lateral connection and GANs | |
| CN109544492A (en) | A kind of multi-focus image fusion data set production method based on convolutional neural networks | |
| CN112862849B (en) | Image segmentation and full convolution neural network-based field rice ear counting method | |
| CN103914683A (en) | Gender identification method and system based on face image | |
| CN109377441B (en) | Tongue image acquisition method and system with privacy protection function | |
| CN107135388A (en) | A kind of depth extraction method of light field image | |
| Wang et al. | Morphological analysis of dendrites and spines by hybridization of ridge detection with twin support vector machine | |
| CN108597003A (en) | A kind of article cover generation method, device, processing server and storage medium | |
| CN110222582A (en) | A kind of image processing method and camera | |
| CN112686258A (en) | Physical examination report information structuring method and device, readable storage medium and terminal | |
| CN109919832A (en) | One kind being used for unpiloted traffic image joining method | |
| Chen et al. | A fusion-based defogging algorithm | |
| Zhang et al. | Removing Foreground Occlusions in Light Field using Micro-lens Dynamic Filter. | |
| Zhang et al. | Residual attentive feature learning network for salient object detection | |
| CN115761574A (en) | Weak surveillance video target segmentation method and device based on frame labeling | |
| CN108520496B (en) | Sea-air background monitoring video image splicing method based on optical flow method | |
| Zhao et al. | A multiple feature-maps interaction pyramid network for defect detection of steel surface | |
| CN114155353A (en) | Point cloud three-dimensional reconstruction method and device based on liquid crystal micro-lens array | |
| Dong et al. | PDDD-PreTrain: a series of commonly used pre-trained models support image-based plant disease diagnosis | |
| CN117765252A (en) | breast cancer identification system and method based on Swin transducer and contrast learning | |
| CN112686247A (en) | Identification card number detection method and device, readable storage medium and terminal | |
| CN110956090A (en) | Face feature extraction method based on positive and negative illumination invariant unit fusion | |
| Liu et al. | Research on driving obstacle detection technology in foggy weather based on GCANet and feature fusion training | |
| Jiang et al. | RDD-YOLOv5: Road Defect Detection Algorithm with Self-Attention Based on Unmanned Aerial Vehicle Inspection |
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: 20190329 |
|
| RJ01 | Rejection of invention patent application after publication |