+ All Categories
Home > Documents > Chapter 8

Chapter 8

Date post: 22-Jan-2016
Category:
Upload: dinh
View: 32 times
Download: 0 times
Share this document with a friend
Description:
Chapter 8. Documented applications of TRS and affine moment invariants. Character/digit/symbol recognition Recognition of aircraft and ship silhouettes (also from non-perpendicular views) Recognition of components on an assembly belt - PowerPoint PPT Presentation
Popular Tags:
35
Chapter 8
Transcript
Page 1: Chapter 8

Chapter 8

Page 2: Chapter 8

Documented applications of TRS and affine moment invariants

• Character/digit/symbol recognition • Recognition of aircraft and ship silhouettes

(also from non-perpendicular views)• Recognition of components on an assembly belt• Recognition of biological shapes – algae, fishes, whales, ...• Landmark recognition in robotics• Image registration (medical, satellite, aerial, ...)• Normalization of database images, retriaval• Motion flow estimation• Digital watermarking

Page 3: Chapter 8

Recognition of circular landmarks

Measurement of scoliosis progress during pregnancy

Page 4: Chapter 8

The goal: to detect the landmark centers

The method: template matching by invariants

Page 5: Chapter 8

Recognition of distorted landmarks

Page 6: Chapter 8

Landmark clusters in the space of the AMI’s

Page 7: Chapter 8

Landsat TM SPOT

Satellite image registration

Page 8: Chapter 8

Registration algorithm

• Independent segmentation of both images• Extraction of salient regions• Calculating AMI’s • Finding three most stable pairs in the AMI space• Calculating the primal affine transform parameters• Transforming the SPOT regions over the Landsat• Finding matching regions by minimum distance in the image plane (10 found altogether).

Region centroids serve as final control points• Calculating the final affine transform parameters by a

least-square fit• Resampling and transformation of the SPOT image

Page 9: Chapter 8

Segmentation

Page 10: Chapter 8

Selected regions

Page 11: Chapter 8

Matched region pairs

Page 12: Chapter 8

Matched region pairs

• Three most stable pairs found in the AMI space (the labels in circles) • The other matching regions found by minimum distance in the image plane

Page 13: Chapter 8

Registered and superimposed images

Page 14: Chapter 8

Optical flow estimation

Traditional method A method based on Zernike moments.Note fewer artifacts.

Page 15: Chapter 8

Image retrieval

Moment invariants can be used as features for content-based image retrieval, particularly in case of simple 2D objects.

Page 16: Chapter 8

Digital watermarking by moments

The host image The image with an invisible watermark based on rotation invariants.

Page 17: Chapter 8

Documented applications of convolution and combined invariants

• Character/digit/symbol recognition in the presence of camera shake or other blurs• Robust image registration (medical, satellite)• Camera position estimation through registration• Multichannel deconvolution and superresolution• Detection of image forgeries• Focus/blur measurement

Page 18: Chapter 8

Camera position estimation through registration

Photo at the initial position(sharp)

Photo at the current position,unknown shift and rotation(blurred background because of the object in the foreground)

Page 19: Chapter 8

Position estimation algorithm

• Independent corner detection in both images

• Extraction of salient corner points

• Calculating blur-rotation invariants of a circular neighborhood of each extracted corner

• Matching corners by the invariants (14 matches found)

• Estimating the relative between-image shift and rotation by a least-square fit

Page 20: Chapter 8

Matched corners

Page 21: Chapter 8

Multichannel blind deconvolution

For MBD, robust registration of the input blurred frames is required.

Page 22: Chapter 8

The Poor Fisherman, Paul Gauguin, 1896

MBD of long-exposure images

Page 23: Chapter 8

• Copy-move forgery (clone of a region from the same image)

• The cloned region is often intentionally blurred to make its detection difficult

• Dividing the image into blocks, calculating blur invariants and looking for blocks having the same invariants

• Presence of identical blocks indicates cloning forgery. “Blind” detection without having the original.

Detecting image forgeries

Page 24: Chapter 8

Detecting image forgeries

original

duplicated regions

fake

Page 25: Chapter 8

Recent world-famous photo of Iranian missiles

Page 26: Chapter 8

Duplicated regions indicate that the picture was manipulated

Page 27: Chapter 8

Moment-based focus measure

• Odd-order moments blur invariants

• Even-order moments blur/focus measure

If M(g1) > M(g2) g2 is less blurred

(more focused)

Page 28: Chapter 8

Usage of a focus measure

• Global measurement – ordering a set of images, which differ from each other by a degree of blur, according to their quality. Typically in astronomy.

Page 29: Chapter 8

Images of different level of blur

Page 30: Chapter 8

Sunspots – blurring by atmospheric turbulence

Page 31: Chapter 8

Saturn images – intentional out-of-focus blur

Page 32: Chapter 8

Usage of a focus measure

• Global measurement – ordering a set of images, which differ from each other by a degree of blur, according to their quality. Typically in astronomy.

• The moments perform very well in the above cases because of their robustness to noise.

Page 33: Chapter 8

Usage of a focus measure

• Local measurement – selecting the frame in which a certain small region is sharp/least defocussed. Typically in multifocal image fusion.

Page 34: Chapter 8

Multifocus fusion based on a localblur measurement

Page 35: Chapter 8

Usage of a focus measure

• Local measurement – selecting the frame in which a certain small region is sharp/least defocussed. Typically in multifocal image fusion.

• The moments are worse than wavelets and Laplacian because of their global character.


Recommended