SlideShare a Scribd company logo
1 of 1
Download to read offline
Photo Synthesis: Image Stitching and Registration
Keenan Johnstone, Travis Gray, Dr. Mark Eramian, Dr. Kevin Stanley, Dr. Ian Stavness
Problem
Agricultural researchers are beginning to use aerial drones to take multi-spectral
images (images that focus on a narrow spectrum of light) of their crops. However,
doing so results in some unusual challenges.
• The cameras that take images at the different spectrums usually have a
different lense for each specrum.
• The different cameras also don't all capture at the same time.
Together, this means that the images it take will all be slightly misaligned.
Registration vs. Stitching
SolutionTo compensate for this misalignment, we needed to register the images
across the spectrums. That is, we needed to line up the different images so
that the plants are in the same place across all images.
• Fourier transforms were used to take the multiple images and calculate
the optimal phase shift to align the images.
• The results were aligned, but there were several cases of blurred edges
around certain parts of the image.
• Due to this error, we used a more accurate method in which keypoints
in the image were found and matched between the images.
• This allowed us to create a homography between spectrums
(A 3x3 matrix that is an instruction set on how to scale, rotate, and
warp a set of pixels).
• With these matrices, we found much better results, where images were
aligned with subpixel accuracy.
P
2
IRC
This multispectral camera is actually made up of 5 separate cameras, each of which take a shot at a slightly
different time. Red, Green, Blue, Near-Infared, and Rededge.
Images taken in the different spectrums:
Blue, Green, Red, Near Infared, and Rededge respectively
The resulting image after using a homography on the blue and red specturms to align with the
green scpectrum. As you can see from the outside edges, the blue spectrum
had to be shifted to the right by quite a bit, and the red had to be shifted up.
Another application of this key point matching is the ability to stitch together images.
Drones have to take multiple, separate images of a field to be able to get the entire
thing in decent quality. This means that many smaller images will need to be stitched
together to create a large composite image of the entire field.
This can be done by the same keypoint matching as before, but now applied to find
where the overlap of smaller images are, and combine those points to create a bigger
image from the multiple.
These two images show where the matched key points exist in images where some overlap occer.
The bottom of the left image also appears in the top of the right image.
These two images, now stitched together.
The homography relating the right image to the left image.
The result of stitching multiple images together
Constructing a system that can register and stitch drone images has
streamlined the process of plant image analysis by reducing human input.
This is important because all image analysis methods on plants require
aligned colour channels and a complete view of the region of interest.
Future improvements include accounting for discolouration during
registration, and applying orthorectification techniques to the drone
images before stitching.
Conclusion

More Related Content

What's hot

Remote Sensing: Principal Component Analysis
Remote Sensing: Principal Component AnalysisRemote Sensing: Principal Component Analysis
Remote Sensing: Principal Component AnalysisKamlesh Kumar
 
Math 390 - Machine Learning Techniques Presentation
Math 390 - Machine Learning Techniques PresentationMath 390 - Machine Learning Techniques Presentation
Math 390 - Machine Learning Techniques PresentationDarragh Punch
 
Week3pres sample
Week3pres sampleWeek3pres sample
Week3pres sampleed136
 
Lighting simulation i
Lighting simulation iLighting simulation i
Lighting simulation iNguyen Khuong
 
Inter annual insolation variability (solar resource)
Inter annual insolation variability (solar resource)Inter annual insolation variability (solar resource)
Inter annual insolation variability (solar resource)chaudharichetan
 

What's hot (8)

Remote Sensing: Principal Component Analysis
Remote Sensing: Principal Component AnalysisRemote Sensing: Principal Component Analysis
Remote Sensing: Principal Component Analysis
 
SRJC Preso Final
SRJC Preso FinalSRJC Preso Final
SRJC Preso Final
 
Math 390 - Machine Learning Techniques Presentation
Math 390 - Machine Learning Techniques PresentationMath 390 - Machine Learning Techniques Presentation
Math 390 - Machine Learning Techniques Presentation
 
Week3pres sample
Week3pres sampleWeek3pres sample
Week3pres sample
 
Seasons Interactive
Seasons InteractiveSeasons Interactive
Seasons Interactive
 
Lighting plan
Lighting planLighting plan
Lighting plan
 
Lighting simulation i
Lighting simulation iLighting simulation i
Lighting simulation i
 
Inter annual insolation variability (solar resource)
Inter annual insolation variability (solar resource)Inter annual insolation variability (solar resource)
Inter annual insolation variability (solar resource)
 

Similar to Stitching-Registration-Poster

AE 497 Spring 2015 Final Report
AE 497 Spring 2015 Final ReportAE 497 Spring 2015 Final Report
AE 497 Spring 2015 Final ReportCatherine McCarthy
 
A Novel Color Image Fusion for Multi Sensor Night Vision Images
A Novel Color Image Fusion for Multi Sensor Night Vision ImagesA Novel Color Image Fusion for Multi Sensor Night Vision Images
A Novel Color Image Fusion for Multi Sensor Night Vision ImagesEditor IJCATR
 
Photography
PhotographyPhotography
Photographysydney20
 
Fusing stereo images into its equivalent cyclopean view
Fusing stereo images into its equivalent cyclopean viewFusing stereo images into its equivalent cyclopean view
Fusing stereo images into its equivalent cyclopean viewEngineering Publication House
 
Solar Radiation Estimation based on Digital Image Processing
Solar Radiation Estimation based on Digital Image ProcessingSolar Radiation Estimation based on Digital Image Processing
Solar Radiation Estimation based on Digital Image ProcessingPrashant Pal
 
Remote Sensing: Resolution Merge
Remote Sensing: Resolution MergeRemote Sensing: Resolution Merge
Remote Sensing: Resolution MergeKamlesh Kumar
 
RADAR Image Fusion Using Wavelet Transform
RADAR Image Fusion Using Wavelet TransformRADAR Image Fusion Using Wavelet Transform
RADAR Image Fusion Using Wavelet TransformINFOGAIN PUBLICATION
 
An Unsupervised Change Detection in Satellite IMAGES Using MRFFCM Clustering
An Unsupervised Change Detection in Satellite IMAGES Using MRFFCM ClusteringAn Unsupervised Change Detection in Satellite IMAGES Using MRFFCM Clustering
An Unsupervised Change Detection in Satellite IMAGES Using MRFFCM ClusteringEditor IJCATR
 
image-processing-husseina-ozigi-otaru.ppt
image-processing-husseina-ozigi-otaru.pptimage-processing-husseina-ozigi-otaru.ppt
image-processing-husseina-ozigi-otaru.pptRaviSharma65345
 
Motivation for image fusion
Motivation for image fusionMotivation for image fusion
Motivation for image fusionVIVEKANAND BONAL
 
Efficient 3D stereo vision stabilization for multi-camera viewpoints
Efficient 3D stereo vision stabilization for multi-camera viewpointsEfficient 3D stereo vision stabilization for multi-camera viewpoints
Efficient 3D stereo vision stabilization for multi-camera viewpointsjournalBEEI
 
Rules of composition
Rules of compositionRules of composition
Rules of compositionChronoPinoyX
 
Coneal topography instrumentation, techniques, procedures, limitations, advan...
Coneal topography instrumentation, techniques, procedures, limitations, advan...Coneal topography instrumentation, techniques, procedures, limitations, advan...
Coneal topography instrumentation, techniques, procedures, limitations, advan...Raju Kaiti
 
Lecture Vray Rendering
Lecture Vray RenderingLecture Vray Rendering
Lecture Vray RenderingMuddasirShah3
 
5. lecture 4 data capturing techniques - satellite and aerial images
5. lecture 4   data capturing techniques - satellite and aerial images5. lecture 4   data capturing techniques - satellite and aerial images
5. lecture 4 data capturing techniques - satellite and aerial imagesFenTaHun6
 

Similar to Stitching-Registration-Poster (20)

AE 497 Spring 2015 Final Report
AE 497 Spring 2015 Final ReportAE 497 Spring 2015 Final Report
AE 497 Spring 2015 Final Report
 
A Novel Color Image Fusion for Multi Sensor Night Vision Images
A Novel Color Image Fusion for Multi Sensor Night Vision ImagesA Novel Color Image Fusion for Multi Sensor Night Vision Images
A Novel Color Image Fusion for Multi Sensor Night Vision Images
 
Photography Terminology PowerPoint
Photography Terminology PowerPointPhotography Terminology PowerPoint
Photography Terminology PowerPoint
 
Aberration_Errors
Aberration_ErrorsAberration_Errors
Aberration_Errors
 
Photography
PhotographyPhotography
Photography
 
Fusing stereo images into its equivalent cyclopean view
Fusing stereo images into its equivalent cyclopean viewFusing stereo images into its equivalent cyclopean view
Fusing stereo images into its equivalent cyclopean view
 
Solar Radiation Estimation based on Digital Image Processing
Solar Radiation Estimation based on Digital Image ProcessingSolar Radiation Estimation based on Digital Image Processing
Solar Radiation Estimation based on Digital Image Processing
 
Remote Sensing: Resolution Merge
Remote Sensing: Resolution MergeRemote Sensing: Resolution Merge
Remote Sensing: Resolution Merge
 
RADAR Image Fusion Using Wavelet Transform
RADAR Image Fusion Using Wavelet TransformRADAR Image Fusion Using Wavelet Transform
RADAR Image Fusion Using Wavelet Transform
 
Mapping
MappingMapping
Mapping
 
An Unsupervised Change Detection in Satellite IMAGES Using MRFFCM Clustering
An Unsupervised Change Detection in Satellite IMAGES Using MRFFCM ClusteringAn Unsupervised Change Detection in Satellite IMAGES Using MRFFCM Clustering
An Unsupervised Change Detection in Satellite IMAGES Using MRFFCM Clustering
 
image-processing-husseina-ozigi-otaru.ppt
image-processing-husseina-ozigi-otaru.pptimage-processing-husseina-ozigi-otaru.ppt
image-processing-husseina-ozigi-otaru.ppt
 
Dd25624627
Dd25624627Dd25624627
Dd25624627
 
Motivation for image fusion
Motivation for image fusionMotivation for image fusion
Motivation for image fusion
 
Efficient 3D stereo vision stabilization for multi-camera viewpoints
Efficient 3D stereo vision stabilization for multi-camera viewpointsEfficient 3D stereo vision stabilization for multi-camera viewpoints
Efficient 3D stereo vision stabilization for multi-camera viewpoints
 
Rules of composition
Rules of compositionRules of composition
Rules of composition
 
Coneal topography instrumentation, techniques, procedures, limitations, advan...
Coneal topography instrumentation, techniques, procedures, limitations, advan...Coneal topography instrumentation, techniques, procedures, limitations, advan...
Coneal topography instrumentation, techniques, procedures, limitations, advan...
 
symfeat_cvpr2012.pdf
symfeat_cvpr2012.pdfsymfeat_cvpr2012.pdf
symfeat_cvpr2012.pdf
 
Lecture Vray Rendering
Lecture Vray RenderingLecture Vray Rendering
Lecture Vray Rendering
 
5. lecture 4 data capturing techniques - satellite and aerial images
5. lecture 4   data capturing techniques - satellite and aerial images5. lecture 4   data capturing techniques - satellite and aerial images
5. lecture 4 data capturing techniques - satellite and aerial images
 

Stitching-Registration-Poster

  • 1. Photo Synthesis: Image Stitching and Registration Keenan Johnstone, Travis Gray, Dr. Mark Eramian, Dr. Kevin Stanley, Dr. Ian Stavness Problem Agricultural researchers are beginning to use aerial drones to take multi-spectral images (images that focus on a narrow spectrum of light) of their crops. However, doing so results in some unusual challenges. • The cameras that take images at the different spectrums usually have a different lense for each specrum. • The different cameras also don't all capture at the same time. Together, this means that the images it take will all be slightly misaligned. Registration vs. Stitching SolutionTo compensate for this misalignment, we needed to register the images across the spectrums. That is, we needed to line up the different images so that the plants are in the same place across all images. • Fourier transforms were used to take the multiple images and calculate the optimal phase shift to align the images. • The results were aligned, but there were several cases of blurred edges around certain parts of the image. • Due to this error, we used a more accurate method in which keypoints in the image were found and matched between the images. • This allowed us to create a homography between spectrums (A 3x3 matrix that is an instruction set on how to scale, rotate, and warp a set of pixels). • With these matrices, we found much better results, where images were aligned with subpixel accuracy. P 2 IRC This multispectral camera is actually made up of 5 separate cameras, each of which take a shot at a slightly different time. Red, Green, Blue, Near-Infared, and Rededge. Images taken in the different spectrums: Blue, Green, Red, Near Infared, and Rededge respectively The resulting image after using a homography on the blue and red specturms to align with the green scpectrum. As you can see from the outside edges, the blue spectrum had to be shifted to the right by quite a bit, and the red had to be shifted up. Another application of this key point matching is the ability to stitch together images. Drones have to take multiple, separate images of a field to be able to get the entire thing in decent quality. This means that many smaller images will need to be stitched together to create a large composite image of the entire field. This can be done by the same keypoint matching as before, but now applied to find where the overlap of smaller images are, and combine those points to create a bigger image from the multiple. These two images show where the matched key points exist in images where some overlap occer. The bottom of the left image also appears in the top of the right image. These two images, now stitched together. The homography relating the right image to the left image. The result of stitching multiple images together Constructing a system that can register and stitch drone images has streamlined the process of plant image analysis by reducing human input. This is important because all image analysis methods on plants require aligned colour channels and a complete view of the region of interest. Future improvements include accounting for discolouration during registration, and applying orthorectification techniques to the drone images before stitching. Conclusion