Rotating Half Smoothing Filters, Image Rotating Half Smoothing - - PowerPoint PPT Presentation

▶
rotating half smoothing filters image rotating half
SMART_READER_LITE
LIVE PREVIEW

Rotating Half Smoothing Filters, Image Rotating Half Smoothing - - PowerPoint PPT Presentation

Jan 08, 2023 197 likes •852 views

Rotating Half Smoothing Filters, Image Rotating Half Smoothing Filters, Image Segmentation and Anisotropic Diffusion Segmentation and Anisotropic Diffusion Baptiste Magnier Baptiste Magnier http://www.lgi2p.ema.fr/~magnier/

slide-1
SLIDE 1

1/43

Rotating Half Smoothing Filters, Image Rotating Half Smoothing Filters, Image Segmentation and Anisotropic Diffusion Segmentation and Anisotropic Diffusion

Baptiste Magnier Baptiste Magnier

http://www.lgi2p.ema.fr/~magnier/ http://www.lgi2p.ema.fr/~magnier/

Journée Des Doctorants du LGi2P 2011 Journée Des Doctorants du LGi2P 2011 Nîmes, 28/06/2011 Nîmes, 28/06/2011

slide-2
SLIDE 2

Daniel Diep received an engineer degree in Electrical Engineering and a PhD in Automatic Control. He works on modelling and control of distributed systems, neural networks, multi-agent systems, various applications in the domains of signal processing, manufacturing systems, and more recently in image processing.

  • A. Lueder, J. Peschke, T. Sauter, S. Deter, D. Diep :

"Distributed intelligence for plant automation based on multi-agent systems : the PABADIS approach", in Production Planning & Control, Vol 15 N°2, 2004, 201-212.

  • D. Diep, C. Alexakos, C. Wagner :

"An Ontology-based Interoperability Framework for Distributed Manufacturing Control". 12th IEEE Conference on Emerging Technologies and Factory Automation (ETFA), Greece, 2007.

  • K. Benaissa, D. Diep, A. Dolgui:

"Control of chaos in agent based manufacturing systems". 13th IEEE International Conference on Emerging Technologies and Factory Automation (ETFA), Hamburg, 2008, pp. 1252-1259. Philippe Montesinos works in image processing: image segmentation, motion analysis, matching points, diffusion...

  • N. Armande, P. Montesinos, O. Monga.

A 3D Thin Nets Extraction Method for Medical Imaging. In Proceedings of ICPR, Vol. 1, Track A, pp. 642-646, Vienna, 1996.

  • V. Gouet, P. Montesinos, D. Pelé.

A Fast Matching Method for Color Uncalibrated Images Using Differential Invariants. In BMVC-98, The Ninth British Machine Vision Conference. Southampton UK, pp 14-17, 1998.

  • D. Sidibe, P. Montesinos and S. Janaqi.

Matching Local Invariant Features with Contextual Information: An Experimental Evaluation. Electronic Letters on Computer Vision and Image Analysis, 7(1):26-39, 2008.

  • P. Montesinos, B. Magnier.

A New Perceptual Edge Detector in Color Images. In Advanced Concepts for Intelligent Vision Systems, ACIVS 2010, Sydney, Australia.

slide-3
SLIDE 3

Rotating Half Smoothing Filters, Image Rotating Half Smoothing Filters, Image Segmentation and Anisotropic Diffusion Segmentation and Anisotropic Diffusion

  • Anisotropic edge detection
  • Anisotropic edge detection
  • Crest and Valley extraction
  • Crest and Valley extraction
  • Texture Removal by Anisotropic Diffusion
  • Texture Removal by Anisotropic Diffusion
slide-4
SLIDE 4

Rotating Half Smoothing Filters, Image Rotating Half Smoothing Filters, Image Segmentation and Anisotropic Diffusion Segmentation and Anisotropic Diffusion

  • Anisotropic edge detection
  • Anisotropic edge detection
  • Crest and Valley extraction
  • Crest and Valley extraction
  • Texture Removal by Anisotropic Diffusion
  • Texture Removal by Anisotropic Diffusion
slide-5
SLIDE 5

Rotating Half Smoothing Filters, Image Rotating Half Smoothing Filters, Image Segmentation and Anisotropic Diffusion Segmentation and Anisotropic Diffusion

  • Anisotropic edge detection
  • Anisotropic edge detection
  • Crest and Valley extraction
  • Crest and Valley extraction
  • Texture Removal by Anisotropic Diffusion
  • Texture Removal by Anisotropic Diffusion
slide-6
SLIDE 6

Rotating Half Smoothing Filters, Image Rotating Half Smoothing Filters, Image Segmentation and Anisotropic Diffusion Segmentation and Anisotropic Diffusion

  • Anisotropic edge detection
  • Anisotropic edge detection
  • Crest and Valley extraction
  • Crest and Valley extraction
  • Texture Removal by Anisotropic Diffusion
  • Texture Removal by Anisotropic Diffusion

Philippe Montesinos and Baptiste Magnier A New Perceptual Edge Detector in Color Images. In Advanced Concepts for Intelligent Vision Systems 2010 (ACIVS 2010), 2010, Sydney, Australia

slide-7
SLIDE 7

What are edges in an image? What are edges in an image?

  • Edges correspond to object boundaries
  • Pixels where image brightness changes significantly
  • Edge extraction : Contours are calculated from image function behavior

in the neighborhood of the pixel

slide-8
SLIDE 8

Edge detection - overview Edge detection - overview

Image Smoothing / Regularization Gradient & Direction

Edges extracted by computing local maxima of the gradient in the gradient direction

Hysteresis threshold Binarisation

slide-9
SLIDE 9

Anisotropic edge detection Anisotropic edge detection

Our anisotropic edge detector is based on the use of two elongated and oriented Filters in two different directions

slide-10
SLIDE 10

Anisotropic edge detection Anisotropic edge detection

Derivative Filter : Derivative Filter :

slide-11
SLIDE 11

Anisotropic edge detection Anisotropic edge detection

slide-12
SLIDE 12

Anisotropic edge detection Anisotropic edge detection

Positive or negative peaks correspond to directions of contours Positive or negative peaks correspond to directions of contours

slide-13
SLIDE 13

Anisotropic edge detection Anisotropic edge detection

slide-14
SLIDE 14

Results Results

slide-15
SLIDE 15

Results Results

Canny Our result

slide-16
SLIDE 16

Results Results

slide-17
SLIDE 17

Perceptual Result Perceptual Result

slide-18
SLIDE 18

Perceptual Result Perceptual Result

slide-19
SLIDE 19

Anisotropic Edge Detection Using Anisotropic Edge Detection Using Gamma Correction in Color Images : Gamma Correction in Color Images : ANEG ANEG

Baptiste Magnier, Philippe Montesinos and Daniel Diep Fast Anisotropic Edge Detection Using Gamma Correction in Color Images. In IEEE 7th International Symposium on Image and Signal Processing and Analysis (ISPA 2011), September 4-6, 2011, Dubrovnik, Croatia

slide-20
SLIDE 20

Gamma Correction Gamma Correction

slide-21
SLIDE 21
slide-22
SLIDE 22

Rotating Half Smoothing Filters, Image Rotating Half Smoothing Filters, Image Segmentation and Anisotropic Diffusion Segmentation and Anisotropic Diffusion

Baptiste Magnier, Philippe Montesinos and Daniel Diep. Ridges and Valleys Detection in Images using Difference of Rotating Half Smoothing Filters. In Advanced Concepts for Intelligent Vision Systems 2011 (ACIVS 2011), August 22-25, 2011, Ghent, Belgium.

  • Anisotropic edge detection
  • Anisotropic edge detection
  • Crest and Valley extraction
  • Crest and Valley extraction
  • Texture Removal by Anisotropic Diffusion
  • Texture Removal by Anisotropic Diffusion
slide-23
SLIDE 23

The problem The problem

?

slide-24
SLIDE 24

What is a crest line ? What is a crest line ?

Crest line, ridge or valley are roof edges Crest line, ridge or valley are roof edges

Edge detection on crest line → two different edges on both sides of the crest Edge detection on crest line → two different edges on both sides of the crest

Edge detection Ridge White line Valley Black line

slide-25
SLIDE 25

Perceptual curve Perceptual curve

Anisotropic edge detectors on discontinuous roof edges Anisotropic edge detectors on discontinuous roof edges

Edge detection Anisotropic Edge detection

Baptiste Magnier, Daniel Diep and Philippe Montesinos : Perceptual Curve Extraction. In The 10th IEEE IVMSP (Image, Video, and Multidimensional Signal Processing Technical Committee) on "Perception and Visual Signal Analysis", June 16-17, 2011, Ithaca, USA.

slide-26
SLIDE 26

A new curve detector which involves anisotropic directional linear Filtering by means of difference of two half rotating Gaussian Filters (DRF).

ANISOTROPIC CURVE EXTRACTION ANISOTROPIC CURVE EXTRACTION

We have computed this ridge/valley

  • perator

using a local directional maximization/minimization of the Filters response. For a pixel belonging to a crest line, it corresponds to an entering and leaving path : two half Filters.

slide-27
SLIDE 27

Difference of Rotated Half Smoothing Filters (DRF) Difference of Rotated Half Smoothing Filters (DRF)

D D is an anisotropic DoG using two filters at the same is an anisotropic DoG using two filters at the same

  • rientation, same height but different widths :
  • rientation, same height but different widths :

Discretized DRF

slide-28
SLIDE 28

Difference of images smoothed at the same orientation using Difference of images smoothed at the same orientation using 2 different filters → a bank of DoG images 2 different filters → a bank of DoG images

Difference of Rotated Half Smoothing Filters (DRF) Difference of Rotated Half Smoothing Filters (DRF)

slide-29
SLIDE 29

Peaks correspond to directions of crest lines

slide-30
SLIDE 30
slide-31
SLIDE 31

Ridges and Valleys Extraction Ridges and Valleys Extraction

slide-32
SLIDE 32

Ridges and Valleys Extraction Ridges and Valleys Extraction

slide-33
SLIDE 33

Ridges and Valleys Extraction Ridges and Valleys Extraction

slide-34
SLIDE 34

DRF Results DRF Results

slide-35
SLIDE 35
slide-36
SLIDE 36
slide-37
SLIDE 37
slide-38
SLIDE 38
slide-39
SLIDE 39
slide-40
SLIDE 40
slide-41
SLIDE 41
slide-42
SLIDE 42

Pine trunk cutting Valleys Ridges

slide-43
SLIDE 43
slide-44
SLIDE 44

Ridge And Valley Junctions Extraction Ridge And Valley Junctions Extraction

Baptiste Magnier, Philippe Montesinos and Daniel Diep. Ridge and Valley Junctions Extraction. In The 2011 International Conference on Image Processing, Computer Vision, & Pattern Recognition (IPCV'11), July 18-21, 2011, Las Vegas, USA.

slide-45
SLIDE 45

Ridge And Valley Junctions Extraction Ridge And Valley Junctions Extraction

For each pixel, we compute J the sum of the 4 higher positive peaks and the 4 lower negative peaks followed by a spatial local maximum in the image of J for ridge junctions (and the spatial local minima for valleys).

slide-46
SLIDE 46
slide-47
SLIDE 47
slide-48
SLIDE 48

Rotating Half Smoothing Filters, Image Rotating Half Smoothing Filters, Image Segmentation and Anisotropic Diffusion Segmentation and Anisotropic Diffusion

  • Anisotropic edge detection
  • Anisotropic edge detection
  • Crest and Valley extraction
  • Crest and Valley extraction
  • Texture Removal by Anisotropic Diffusion
  • Texture Removal by Anisotropic Diffusion

Baptiste Magnier, Philippe Montesinos and Daniel Diep. Texture Removal by Pixel Classification using a Rotating Filter. In IEEE 36th International Conference on Acoustics, Speech and Signal Processing (ICASSP 2011), May 2011, Prague, Czech Republic.

slide-49
SLIDE 49

Original Image segmentation

slide-50
SLIDE 50

Original Isotropic diffusion 50 iterations

slide-51
SLIDE 51

Rotating half Gaussian smoothing Filters

slide-52
SLIDE 52

Rotating half Gaussian smoothing Filters

Image convolution with each filter [0,360] → circular signal attached to each pixel → function of the angle

slide-53
SLIDE 53

Rotating half Gaussian smoothing Filters

slide-54
SLIDE 54

Rotating half Gaussian smoothing Filters

slide-55
SLIDE 55

Pixel signal : cartesian representation

Point 8 : texture Point 1 : edge

slide-56
SLIDE 56

The directions of maximum (or minimum) curvature delimit the flat area detected These directions will be used by the anisotropic diffusion

Flat area detection

Original signal

The method detects significant flat areas, which correspond to homogeneous regions → We are able to classify if a pixel belongs to

  • A texture
  • A homogenous region

→ Isotropic case

  • A contour
  • A point between a homogenous region and

a texture → Anisotropic case

slide-57
SLIDE 57

Flat area detection

slide-58
SLIDE 58

Our result 50 iterations Original

slide-59
SLIDE 59

Original image 420x395 MCM 50 iterations PM 100 iterations Tschumperlé 50 iterations Alvarez 100 iterations Our method, 50 iterations

Fiber effect Texture is considered as edge Rough edge localisation → removes texture and edges → removes texture only

slide-60
SLIDE 60

Original image 420x395 MCM 50 iterations PM 100 iterations Tschumperlé 50 iterations Tschumperlé 50 iterations Our method, 50 iterations

Fiber effect Blur edges → busches Rough edge localisation → removes texture only

slide-61
SLIDE 61

Edge detection after the smoothing process

Our result Original Alvarez Tschumperlé

slide-62
SLIDE 62
  • Texture Removal by Anisotropic Diffusion
  • Texture Removal by Anisotropic Diffusion

→ Extension to color images → Extension to color images

Baptiste Magnier, Philippe Montesinos and Daniel Diep. Texture Removal in Color Images by Anisotropic Diffusion. In International Conference on Computer Vision Theory and Applications (VISAPP), 2011, Algarve, Portugal.

slide-63
SLIDE 63

Conclusion Conclusion

A new anisotropic Gaussian filter based on the use of two elongated and A new anisotropic Gaussian filter based on the use of two elongated and

  • riented filters in two different directions.
  • riented filters in two different directions.

Using these rotating kernels allows to : Using these rotating kernels allows to : Extract edges ( → derivative filter) Extract edges ( → derivative filter) Detect crest lines ( → DoG filter) Detect crest lines ( → DoG filter) Diffuse anisotropically textures ( → smoothing filter) Diffuse anisotropically textures ( → smoothing filter) * ++++++++++ * * ++++++++++ *

Future Works Future Works

Anisotropic diffusion : Image restoration, Medical Images... Anisotropic diffusion : Image restoration, Medical Images... Crest lines : Medical images (blood vessels...) Crest lines : Medical images (blood vessels...) Corners and junctions detection using the two directions of the kernels. Corners and junctions detection using the two directions of the kernels. Potential applications : camera calibration and interest points matching. Potential applications : camera calibration and interest points matching.

slide-64
SLIDE 64
slide-65
SLIDE 65

Philippe Montesinos and Baptiste Magnier : A New Perceptual Edge Detector in Color Images. In Advanced Concepts for Intelligent Vision Systems 2010 (ACIVS 2010), 2010, Sydney, Australia Baptiste Magnier, Philippe Montesinos and Daniel Diep : Texture Removal by Pixel Classification using a Rotating Filter. In IEEE 36th International Conference on Acoustics, Speech and Signal Processing (ICASSP 2011), May 2011, Prague. Baptiste Magnier, Philippe Montesinos and Daniel Diep : Texture Removal in Color Images by Anisotropic Diffusion. In International Conference on Computer Vision Theory and Applications (VISAPP), March 2011, Algarve, Portugal. Baptiste Magnier, Daniel Diep and Philippe Montesinos : Perceptual Curve Extraction. In The 10th IEEE IVMSP (Image, Video, and Multidimensional Signal Processing Technical Committee) on "Perception and Visual Signal Analysis", June 16-17, 2011, Ithaca, USA. Baptiste Magnier, Philippe Montesinos and Daniel Diep : Ridge and Valley Junctions Extraction. In The 2011 International Conference on Image Processing, Computer Vision, & Pattern Recognition (IPCV'11), July 18-21, 2011, Las Vegas, USA. Baptiste Magnier, Philippe Montesinos and Daniel Diep : Ridges and Valleys Detection in Images using Difference of Rotating Half Smoothing Filters. In Advanced Concepts for Intelligent Vision Systems 2011 (ACIVS 2011), August 22-25, 2011, Ghent, Belgium. Baptiste Magnier, Philippe Montesinos and Daniel Diep Fast Anisotropic Edge Detection Using Gamma Correction in Color Images. In 7th IEEE International Symposium on Image and Signal Processing and Analysis (ISPA 2011), September 4-6, 2011, Dubrovnik, Croatia

http://www.lgi2p.ema.fr/~magnier/