Upload
others
View
16
Download
0
Embed Size (px)
Citation preview
Motivation Image Processing Flowchart
Image Acquisition!!
Experimental Results
High Resolution Starry Night Panoramas!Grant Yang!
Department of Electrical Engineering, Stanford University
A B C
.
.
.
.
.
.
Local Registration and Exposure Averaging
.
.
.
Arrows indicate stars which have been recovered using exposure averaging (C) compared to original exposure (A), and noise reduction in Lightroom 5.1 (B)
Noise Reduction:
Camera: Sony A77 Focal Length:16mm Aperture: f2.8 ISO: 3200 Shutter Speed: 15 sec 5 frames with camera rotated in 10 degree increments about the no parallax point 6 exposures/frame
Generating panoramas of the night sky is nontrivial because of • Spatially varying motion of the stars due to the earth’s rotation • Low signal to noise ratio due to low light levels
This project integrates spatially variant image registration into the standard panorama workflow to allow automatic exposure stacking and panorama stitching of the night sky even with the inclusion of foreground elements
Spherical Projection Register Frames Frame Blending
Contrast, Exposure, Sharpness Adjustments
Motion Correction:
A B C
Motion correction (A) results in a substantial improvement in sharpness over the uncorrected image (B). The spatially variant technique results in minimal noticeable artifacts at image boundaries (C)
Final Image
Local Registration Algorithm! • The motion of the stars is modeled as an affine transformation
• The image is segmented into land and sky components via segmentation and morphological operations
• The stars are detected as high intensity peaks in the sky component
• SIFT descriptors and RANSAC are used to calculate the transformation between images
• The image is warped using a weighted sum of the transforms for each region