SlideShare a Scribd company logo
1 of 38
1
Yossi CohenIntro to computer Vision
2
3
How to filter
2d Correlation
h=filter2(g,f); or h=imfilter(f,g);
2d Convolution
h=conv2(g,f);
],[],[],[
,
lnkmflkgnmh
lk
−−= ∑
f=image
g=filter
],[],[],[
,
lnkmflkgnmh
lk
++= ∑
4
Correlation filtering
Say the averaging window size is 2k+1 x 2k+1:
Loop over all pixels in neighborhood around
image pixel F[i,j]
Attribute uniform
weight to each pixel
Now generalize to allow different weights depending on
neighboring pixel’s relative position:
Non-uniform weights
5
Correlation filtering
Filtering an image: replace each pixel with a linear
combination of its neighbors.
The filter “kernel” or “mask” H[u,v] is the prescription for the
weights in the linear combination.
This is called cross-correlation, denoted
6
Properties of smoothing filters
Smoothing
 Values positive
 Sum to 1  constant regions same as input
 Amount of smoothing proportional to mask size
 Remove “high-frequency” components; “low-pass” filter
7
Filtering an impulse signal
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 1 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
a b c
d e f
g h i
What is the result of filtering the impulse signal
(image) F with the arbitrary kernel H?
?
8
Filtering an impulse signal
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 1 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
a b c
d e f
g h i
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 i h g 0 0
0 0 f e d 0 0
0 0 c b a 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
What is the result of filtering the impulse signal
(image) F with the arbitrary kernel H?
9
Averaging filter
 What values belong in the kernel H for the moving
average example?
0 10 20 30 30
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 90 0 90 90 90 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 0 0 0 0 0 0 0
0 0 90 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
111
111
111
“box filter”
?
10
Smoothing by averaging
depicts box filter:
white = high value, black = low value
original filtered
11
Gaussian filter
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 90 0 90 90 90 0 0
0 0 0 90 90 90 90 90 0 0
0 0 0 0 0 0 0 0 0 0
0 0 90 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
1 2 1
2 4 2
1 2 1
This kernel is an
approximation of a 2d
Gaussian function:
12
Smoothing with a Gaussian
13
Smoothing with a Box
14
 Weight contributions of neighboring pixels by nearness
0.003 0.013 0.022 0.013 0.003
0.013 0.059 0.097 0.059 0.013
0.022 0.097 0.159 0.097 0.022
0.013 0.059 0.097 0.059 0.013
0.003 0.013 0.022 0.013 0.003
5 x 5, σ = 1
Slide credit: Christopher Rasmussen
Gaussian
15
Gaussian filters
What parameters matter here?
Size of kernel or mask
Note, Gaussian function has infinite support, but discrete
filters use finite kernels
σ = 5 with
10 x 10
kernel
σ = 5 with
30 x 30
kernel
16
Gaussian filters
What parameters matter here?
Variance of Gaussian: determines extent of
smoothing
σ = 2 with
30 x 30
kernel
σ = 5 with
30 x 30
kernel
17
Matlab
>> hsize = 10;
>> sigma = 5;
>> h = fspecial(‘gaussian’ hsize, sigma);
>> mesh(h);
>> imagesc(h);
>> outim = imfilter(im, h); % correlation
>> imshow(outim);
outim
18
Smoothing with a Gaussian
for sigma=1:3:10
h = fspecial('gaussian‘, fsize, sigma);
out = imfilter(im, h);
imshow(out);
pause;
end
…
Parameter σ is the “scale” / “width” / “spread” of the Gaussian
kernel, and controls the amount of smoothing.
19
Gaussian filters
• Remove “high-frequency” components from the
image (low-pass filter)
Images become more smooth
• Convolution with self is another Gaussian
–So can smooth with small-width kernel, repeat, and
get same result as larger-width kernel would have
–Convolving two times with Gaussian kernel of width σ
is same as convolving once with kernel of width σ√2
• Separable kernel
–Factors into product of two 1D Gaussians
Source: K. Grauman
20
Separability of the Gaussian filter
Source: D. Lowe
21
Separability example
*
*
=
=
2D convolution
(center location only)
Source: K. Grauman
The filter factors
into a product of 1D
filters:
Perform convolution
along rows:
Followed by convolution
along the remaining column:
22
Convolution
 Convolution:
 Flip the filter in both dimensions (bottom to top, right to left)
 Then apply cross-correlation
Notation for
convolution
operator
F
H
23
Convolution vs. correlation
Convolution
Cross-correlation
24
Key properties of linear filters
Linearity:
filter(f1 + f2) = filter(f1) + filter(f2)
Shift invariance: same behavior
regardless of pixel location
filter(shift(f)) = shift(filter(f))
Any linear, shift-invariant operator can be
represented as a convolution
Source: S. Lazebnik
25
More properties
• Commutative: a * b = b * a
 Conceptually no difference between filter and signal
 But particular filtering implementations might break this equality
• Associative: a * (b * c) = (a * b) * c
 Often apply several filters one after another: (((a * b1) * b2) * b3)
 This is equivalent to applying one filter: a * (b1 * b2 * b3)
• Distributes over addition: a * (b + c) = (a * b) + (a * c)
• Scalars factor out: ka * b = a * kb = k (a * b)
• Identity: unit impulse e = [0, 0, 1, 0, 0],
a * e = a Source: S. Lazebnik
26
Matlab
Lab 1 – Smooth/Blur Filters
27
Lets try to blur filter
%%basic image filters
imrgb = imread('peppers.png');
imshow(imrgb);
a = [1 1 1; 1 1 1; 1 1 1]
Corroutimg = filter2(a, imgray);
Convoutimg = conv2(imgray,a);
figure;
imshow(Corroutimg)
figure
imshow(Convoutimg)
28
Fix it
Can we conv2 an RGB image??
Use rgb2gray
Whats wrong now?
Try preserving the image power
Convert image to double
29
Results
%%basic image filters
imrgb = imread('peppers.png');
imgray =
im2double(rgb2gray(imrgb));
imshow(imgray);
a = 1/16*ones(4)
Corroutimg = filter2(a, imgray);
Convoutimg = conv2(a, imgray);
figure;
imshow(Corroutimg)
figure
imshow(Convoutimg)
30
Sharp
%%basic sharp filter
imrgb = imread('peppers.png');
imgray =
im2double(rgb2gray(imrgb));
imshow(imgray);
a = [ 0 0 0; 0 2 0; 0 0 0] -
1/9*ones(3)
Corroutimg = filter2(a,
imgray);
figure;
imshow(Corroutimg)
31
Practical matters
What happens near the edge?
the filter window falls off the edge of the image
need to extrapolate
methods:
clip filter (black)
wrap around
copy edge
reflect across edge
Source: S. Marschner
32
Matlab edge aware filtering
methods (MATLAB):
clip filter (black): imfilter(f, g, 0)
wrap around: imfilter(f, g, ‘circular’)
copy edge: imfilter(f, g, ‘replicate’)
reflect across edge: imfilter(f, g, ‘symmetric’)
33
Practical matters
What is the size of the output?
• MATLAB: filter2(g, f, shape)
shape = ‘full’: output size is sum of sizes of f and g
shape = ‘same’: output size is same as f
shape = ‘valid’: output size is difference of sizes of f and g
f
gg
gg
f
gg
gg
f
gg
gg
full same valid
34
Median filters
A Median Filter operates over a window by
selecting the median intensity in the window.
What advantage does a median filter have
over a mean filter?
Is a median filter a kind of convolution?
35
Matlab
Lab 2 – Median Filter
36
Comparison: salt and pepper noise
37
Median Filter Example
%%Median
imrgb = imread('peppers.png');
imgray =
im2double(rgb2gray(imrgb));
imnoise = imnoise(imgray,'salt &
pepper',0.1);
imshow(imnoise)
figure
imshow(medfilt2(imnoise))
38
Basic Filters summary
Linear filtering is sum of dot
product at each position
Can smooth, sharpen, translate
(among many other uses)
Be aware of details for filter size,
extrapolation, cropping
111
111
111

More Related Content

What's hot

Image segmentation ppt
Image segmentation pptImage segmentation ppt
Image segmentation ppt
Gichelle Amon
 

What's hot (20)

Noise
NoiseNoise
Noise
 
Sharpening spatial filters
Sharpening spatial filtersSharpening spatial filters
Sharpening spatial filters
 
Chapter10 image segmentation
Chapter10 image segmentationChapter10 image segmentation
Chapter10 image segmentation
 
Digital image processing
Digital image processing  Digital image processing
Digital image processing
 
Image enhancement
Image enhancementImage enhancement
Image enhancement
 
Lect 02 first portion
Lect 02   first portionLect 02   first portion
Lect 02 first portion
 
Image segmentation ppt
Image segmentation pptImage segmentation ppt
Image segmentation ppt
 
Spatial Filters (Digital Image Processing)
Spatial Filters (Digital Image Processing)Spatial Filters (Digital Image Processing)
Spatial Filters (Digital Image Processing)
 
Image segmentation
Image segmentationImage segmentation
Image segmentation
 
Sharpening using frequency Domain Filter
Sharpening using frequency Domain FilterSharpening using frequency Domain Filter
Sharpening using frequency Domain Filter
 
Image Representation & Descriptors
Image Representation & DescriptorsImage Representation & Descriptors
Image Representation & Descriptors
 
IMAGE SEGMENTATION.
IMAGE SEGMENTATION.IMAGE SEGMENTATION.
IMAGE SEGMENTATION.
 
Histogram Processing
Histogram ProcessingHistogram Processing
Histogram Processing
 
Digital image processing- Compression- Different Coding techniques
Digital image processing- Compression- Different Coding techniques Digital image processing- Compression- Different Coding techniques
Digital image processing- Compression- Different Coding techniques
 
SIFT
SIFTSIFT
SIFT
 
Image feature extraction
Image feature extractionImage feature extraction
Image feature extraction
 
06 spatial filtering DIP
06 spatial filtering DIP06 spatial filtering DIP
06 spatial filtering DIP
 
Image Enhancement using Frequency Domain Filters
Image Enhancement using Frequency Domain FiltersImage Enhancement using Frequency Domain Filters
Image Enhancement using Frequency Domain Filters
 
Image Segmentation (Digital Image Processing)
Image Segmentation (Digital Image Processing)Image Segmentation (Digital Image Processing)
Image Segmentation (Digital Image Processing)
 
SPATIAL FILTERING IN IMAGE PROCESSING
SPATIAL FILTERING IN IMAGE PROCESSINGSPATIAL FILTERING IN IMAGE PROCESSING
SPATIAL FILTERING IN IMAGE PROCESSING
 

Similar to Computer Vision - Image Filters

Gaussian filtering 1up
Gaussian filtering 1upGaussian filtering 1up
Gaussian filtering 1up
Pham Hanh
 
05_Spatial_Filtering.ppt
05_Spatial_Filtering.ppt05_Spatial_Filtering.ppt
05_Spatial_Filtering.ppt
pawankamal3
 
ESTIMATING NOISE PARAMETER & FILTERING (Digital Image Processing)
ESTIMATING NOISE PARAMETER & FILTERING (Digital Image Processing)ESTIMATING NOISE PARAMETER & FILTERING (Digital Image Processing)
ESTIMATING NOISE PARAMETER & FILTERING (Digital Image Processing)
Shajun Nisha
 
Enhancement in frequency domain
Enhancement in frequency domainEnhancement in frequency domain
Enhancement in frequency domain
Ashish Kumar
 

Similar to Computer Vision - Image Filters (20)

Gaussian filtering 1up
Gaussian filtering 1upGaussian filtering 1up
Gaussian filtering 1up
 
Spatial Filtering in intro image processingr
Spatial Filtering in intro image processingrSpatial Filtering in intro image processingr
Spatial Filtering in intro image processingr
 
Filtering an image is to apply a convolution
Filtering an image is to apply a convolutionFiltering an image is to apply a convolution
Filtering an image is to apply a convolution
 
2. filtering basics
2. filtering basics2. filtering basics
2. filtering basics
 
03 cie552 image_filtering_spatial
03 cie552 image_filtering_spatial03 cie552 image_filtering_spatial
03 cie552 image_filtering_spatial
 
Image processing 2
Image processing 2Image processing 2
Image processing 2
 
Notes on image processing
Notes on image processingNotes on image processing
Notes on image processing
 
05_Spatial_Filtering.ppt
05_Spatial_Filtering.ppt05_Spatial_Filtering.ppt
05_Spatial_Filtering.ppt
 
ESTIMATING NOISE PARAMETER & FILTERING (Digital Image Processing)
ESTIMATING NOISE PARAMETER & FILTERING (Digital Image Processing)ESTIMATING NOISE PARAMETER & FILTERING (Digital Image Processing)
ESTIMATING NOISE PARAMETER & FILTERING (Digital Image Processing)
 
International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER)International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER)
 
IIR filter design, Digital signal processing
IIR filter design, Digital signal processingIIR filter design, Digital signal processing
IIR filter design, Digital signal processing
 
Enhancement in frequency domain
Enhancement in frequency domainEnhancement in frequency domain
Enhancement in frequency domain
 
Lecture 4
Lecture 4Lecture 4
Lecture 4
 
Lect 03 - first portion
Lect 03 - first portionLect 03 - first portion
Lect 03 - first portion
 
image_enhancement_spatial
 image_enhancement_spatial image_enhancement_spatial
image_enhancement_spatial
 
Wavelet transform in two dimensions
Wavelet transform in two dimensionsWavelet transform in two dimensions
Wavelet transform in two dimensions
 
Image enhancement
Image enhancementImage enhancement
Image enhancement
 
03_definition_bf.ppt
03_definition_bf.ppt03_definition_bf.ppt
03_definition_bf.ppt
 
03_definition_bf [Автосохраненный].ppt
03_definition_bf [Автосохраненный].ppt03_definition_bf [Автосохраненный].ppt
03_definition_bf [Автосохраненный].ppt
 
Spatial filtering
Spatial filteringSpatial filtering
Spatial filtering
 

More from Yoss Cohen

Product wise computer vision development
Product wise computer vision developmentProduct wise computer vision development
Product wise computer vision development
Yoss Cohen
 

More from Yoss Cohen (20)

Infrared simulation and processing on Nvidia platforms
Infrared simulation and processing on Nvidia platformsInfrared simulation and processing on Nvidia platforms
Infrared simulation and processing on Nvidia platforms
 
open platform for swarm training
open platform for swarm training open platform for swarm training
open platform for swarm training
 
Deep Learning - system view
Deep Learning - system viewDeep Learning - system view
Deep Learning - system view
 
Dspip deep learning syllabus
Dspip deep learning syllabusDspip deep learning syllabus
Dspip deep learning syllabus
 
IoT consideration selection
IoT consideration selectionIoT consideration selection
IoT consideration selection
 
IoT evolution
IoT evolutionIoT evolution
IoT evolution
 
Nvidia jetson nano bringup
Nvidia jetson nano bringupNvidia jetson nano bringup
Nvidia jetson nano bringup
 
Autonomous car teleportation architecture
Autonomous car teleportation architectureAutonomous car teleportation architecture
Autonomous car teleportation architecture
 
Motion estimation overview
Motion estimation overviewMotion estimation overview
Motion estimation overview
 
Intro to machine learning with scikit learn
Intro to machine learning with scikit learnIntro to machine learning with scikit learn
Intro to machine learning with scikit learn
 
DASH and HTTP2.0
DASH and HTTP2.0DASH and HTTP2.0
DASH and HTTP2.0
 
HEVC Definitions and high-level syntax
HEVC Definitions and high-level syntaxHEVC Definitions and high-level syntax
HEVC Definitions and high-level syntax
 
Introduction to HEVC
Introduction to HEVCIntroduction to HEVC
Introduction to HEVC
 
FFMPEG on android
FFMPEG on androidFFMPEG on android
FFMPEG on android
 
Hands-on Video Course - "RAW Video"
Hands-on Video Course - "RAW Video" Hands-on Video Course - "RAW Video"
Hands-on Video Course - "RAW Video"
 
Video quality testing
Video quality testingVideo quality testing
Video quality testing
 
HEVC / H265 Hands-On course
HEVC / H265 Hands-On courseHEVC / H265 Hands-On course
HEVC / H265 Hands-On course
 
Web video standards
Web video standardsWeb video standards
Web video standards
 
Product wise computer vision development
Product wise computer vision developmentProduct wise computer vision development
Product wise computer vision development
 
3D Video Programming for Android
3D Video Programming for Android3D Video Programming for Android
3D Video Programming for Android
 

Recently uploaded

Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Safe Software
 
Architecting Cloud Native Applications
Architecting Cloud Native ApplicationsArchitecting Cloud Native Applications
Architecting Cloud Native Applications
WSO2
 
Why Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire businessWhy Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire business
panagenda
 

Recently uploaded (20)

Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
Apidays New York 2024 - The Good, the Bad and the Governed by David O'Neill, ...
 
MS Copilot expands with MS Graph connectors
MS Copilot expands with MS Graph connectorsMS Copilot expands with MS Graph connectors
MS Copilot expands with MS Graph connectors
 
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost SavingRepurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
Repurposing LNG terminals for Hydrogen Ammonia: Feasibility and Cost Saving
 
Apidays Singapore 2024 - Modernizing Securities Finance by Madhu Subbu
Apidays Singapore 2024 - Modernizing Securities Finance by Madhu SubbuApidays Singapore 2024 - Modernizing Securities Finance by Madhu Subbu
Apidays Singapore 2024 - Modernizing Securities Finance by Madhu Subbu
 
Ransomware_Q4_2023. The report. [EN].pdf
Ransomware_Q4_2023. The report. [EN].pdfRansomware_Q4_2023. The report. [EN].pdf
Ransomware_Q4_2023. The report. [EN].pdf
 
AWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of TerraformAWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of Terraform
 
DBX First Quarter 2024 Investor Presentation
DBX First Quarter 2024 Investor PresentationDBX First Quarter 2024 Investor Presentation
DBX First Quarter 2024 Investor Presentation
 
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers:  A Deep Dive into Serverless Spatial Data and FMECloud Frontiers:  A Deep Dive into Serverless Spatial Data and FME
Cloud Frontiers: A Deep Dive into Serverless Spatial Data and FME
 
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
 
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, AdobeApidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
 
MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024
 
Architecting Cloud Native Applications
Architecting Cloud Native ApplicationsArchitecting Cloud Native Applications
Architecting Cloud Native Applications
 
2024: Domino Containers - The Next Step. News from the Domino Container commu...
2024: Domino Containers - The Next Step. News from the Domino Container commu...2024: Domino Containers - The Next Step. News from the Domino Container commu...
2024: Domino Containers - The Next Step. News from the Domino Container commu...
 
FWD Group - Insurer Innovation Award 2024
FWD Group - Insurer Innovation Award 2024FWD Group - Insurer Innovation Award 2024
FWD Group - Insurer Innovation Award 2024
 
AXA XL - Insurer Innovation Award Americas 2024
AXA XL - Insurer Innovation Award Americas 2024AXA XL - Insurer Innovation Award Americas 2024
AXA XL - Insurer Innovation Award Americas 2024
 
Artificial Intelligence Chap.5 : Uncertainty
Artificial Intelligence Chap.5 : UncertaintyArtificial Intelligence Chap.5 : Uncertainty
Artificial Intelligence Chap.5 : Uncertainty
 
A Beginners Guide to Building a RAG App Using Open Source Milvus
A Beginners Guide to Building a RAG App Using Open Source MilvusA Beginners Guide to Building a RAG App Using Open Source Milvus
A Beginners Guide to Building a RAG App Using Open Source Milvus
 
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWEREMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
 
Why Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire businessWhy Teams call analytics are critical to your entire business
Why Teams call analytics are critical to your entire business
 
Apidays Singapore 2024 - Scalable LLM APIs for AI and Generative AI Applicati...
Apidays Singapore 2024 - Scalable LLM APIs for AI and Generative AI Applicati...Apidays Singapore 2024 - Scalable LLM APIs for AI and Generative AI Applicati...
Apidays Singapore 2024 - Scalable LLM APIs for AI and Generative AI Applicati...
 

Computer Vision - Image Filters

  • 1. 1 Yossi CohenIntro to computer Vision
  • 2. 2
  • 3. 3 How to filter 2d Correlation h=filter2(g,f); or h=imfilter(f,g); 2d Convolution h=conv2(g,f); ],[],[],[ , lnkmflkgnmh lk −−= ∑ f=image g=filter ],[],[],[ , lnkmflkgnmh lk ++= ∑
  • 4. 4 Correlation filtering Say the averaging window size is 2k+1 x 2k+1: Loop over all pixels in neighborhood around image pixel F[i,j] Attribute uniform weight to each pixel Now generalize to allow different weights depending on neighboring pixel’s relative position: Non-uniform weights
  • 5. 5 Correlation filtering Filtering an image: replace each pixel with a linear combination of its neighbors. The filter “kernel” or “mask” H[u,v] is the prescription for the weights in the linear combination. This is called cross-correlation, denoted
  • 6. 6 Properties of smoothing filters Smoothing  Values positive  Sum to 1  constant regions same as input  Amount of smoothing proportional to mask size  Remove “high-frequency” components; “low-pass” filter
  • 7. 7 Filtering an impulse signal 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 a b c d e f g h i What is the result of filtering the impulse signal (image) F with the arbitrary kernel H? ?
  • 8. 8 Filtering an impulse signal 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 a b c d e f g h i 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 i h g 0 0 0 0 f e d 0 0 0 0 c b a 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 What is the result of filtering the impulse signal (image) F with the arbitrary kernel H?
  • 9. 9 Averaging filter  What values belong in the kernel H for the moving average example? 0 10 20 30 30 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 90 90 90 90 90 0 0 0 0 0 90 90 90 90 90 0 0 0 0 0 90 90 90 90 90 0 0 0 0 0 90 0 90 90 90 0 0 0 0 0 90 90 90 90 90 0 0 0 0 0 0 0 0 0 0 0 0 0 0 90 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 111 111 111 “box filter” ?
  • 10. 10 Smoothing by averaging depicts box filter: white = high value, black = low value original filtered
  • 11. 11 Gaussian filter 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 90 90 90 90 90 0 0 0 0 0 90 90 90 90 90 0 0 0 0 0 90 90 90 90 90 0 0 0 0 0 90 0 90 90 90 0 0 0 0 0 90 90 90 90 90 0 0 0 0 0 0 0 0 0 0 0 0 0 0 90 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 2 1 2 4 2 1 2 1 This kernel is an approximation of a 2d Gaussian function:
  • 14. 14  Weight contributions of neighboring pixels by nearness 0.003 0.013 0.022 0.013 0.003 0.013 0.059 0.097 0.059 0.013 0.022 0.097 0.159 0.097 0.022 0.013 0.059 0.097 0.059 0.013 0.003 0.013 0.022 0.013 0.003 5 x 5, σ = 1 Slide credit: Christopher Rasmussen Gaussian
  • 15. 15 Gaussian filters What parameters matter here? Size of kernel or mask Note, Gaussian function has infinite support, but discrete filters use finite kernels σ = 5 with 10 x 10 kernel σ = 5 with 30 x 30 kernel
  • 16. 16 Gaussian filters What parameters matter here? Variance of Gaussian: determines extent of smoothing σ = 2 with 30 x 30 kernel σ = 5 with 30 x 30 kernel
  • 17. 17 Matlab >> hsize = 10; >> sigma = 5; >> h = fspecial(‘gaussian’ hsize, sigma); >> mesh(h); >> imagesc(h); >> outim = imfilter(im, h); % correlation >> imshow(outim); outim
  • 18. 18 Smoothing with a Gaussian for sigma=1:3:10 h = fspecial('gaussian‘, fsize, sigma); out = imfilter(im, h); imshow(out); pause; end … Parameter σ is the “scale” / “width” / “spread” of the Gaussian kernel, and controls the amount of smoothing.
  • 19. 19 Gaussian filters • Remove “high-frequency” components from the image (low-pass filter) Images become more smooth • Convolution with self is another Gaussian –So can smooth with small-width kernel, repeat, and get same result as larger-width kernel would have –Convolving two times with Gaussian kernel of width σ is same as convolving once with kernel of width σ√2 • Separable kernel –Factors into product of two 1D Gaussians Source: K. Grauman
  • 20. 20 Separability of the Gaussian filter Source: D. Lowe
  • 21. 21 Separability example * * = = 2D convolution (center location only) Source: K. Grauman The filter factors into a product of 1D filters: Perform convolution along rows: Followed by convolution along the remaining column:
  • 22. 22 Convolution  Convolution:  Flip the filter in both dimensions (bottom to top, right to left)  Then apply cross-correlation Notation for convolution operator F H
  • 24. 24 Key properties of linear filters Linearity: filter(f1 + f2) = filter(f1) + filter(f2) Shift invariance: same behavior regardless of pixel location filter(shift(f)) = shift(filter(f)) Any linear, shift-invariant operator can be represented as a convolution Source: S. Lazebnik
  • 25. 25 More properties • Commutative: a * b = b * a  Conceptually no difference between filter and signal  But particular filtering implementations might break this equality • Associative: a * (b * c) = (a * b) * c  Often apply several filters one after another: (((a * b1) * b2) * b3)  This is equivalent to applying one filter: a * (b1 * b2 * b3) • Distributes over addition: a * (b + c) = (a * b) + (a * c) • Scalars factor out: ka * b = a * kb = k (a * b) • Identity: unit impulse e = [0, 0, 1, 0, 0], a * e = a Source: S. Lazebnik
  • 26. 26 Matlab Lab 1 – Smooth/Blur Filters
  • 27. 27 Lets try to blur filter %%basic image filters imrgb = imread('peppers.png'); imshow(imrgb); a = [1 1 1; 1 1 1; 1 1 1] Corroutimg = filter2(a, imgray); Convoutimg = conv2(imgray,a); figure; imshow(Corroutimg) figure imshow(Convoutimg)
  • 28. 28 Fix it Can we conv2 an RGB image?? Use rgb2gray Whats wrong now? Try preserving the image power Convert image to double
  • 29. 29 Results %%basic image filters imrgb = imread('peppers.png'); imgray = im2double(rgb2gray(imrgb)); imshow(imgray); a = 1/16*ones(4) Corroutimg = filter2(a, imgray); Convoutimg = conv2(a, imgray); figure; imshow(Corroutimg) figure imshow(Convoutimg)
  • 30. 30 Sharp %%basic sharp filter imrgb = imread('peppers.png'); imgray = im2double(rgb2gray(imrgb)); imshow(imgray); a = [ 0 0 0; 0 2 0; 0 0 0] - 1/9*ones(3) Corroutimg = filter2(a, imgray); figure; imshow(Corroutimg)
  • 31. 31 Practical matters What happens near the edge? the filter window falls off the edge of the image need to extrapolate methods: clip filter (black) wrap around copy edge reflect across edge Source: S. Marschner
  • 32. 32 Matlab edge aware filtering methods (MATLAB): clip filter (black): imfilter(f, g, 0) wrap around: imfilter(f, g, ‘circular’) copy edge: imfilter(f, g, ‘replicate’) reflect across edge: imfilter(f, g, ‘symmetric’)
  • 33. 33 Practical matters What is the size of the output? • MATLAB: filter2(g, f, shape) shape = ‘full’: output size is sum of sizes of f and g shape = ‘same’: output size is same as f shape = ‘valid’: output size is difference of sizes of f and g f gg gg f gg gg f gg gg full same valid
  • 34. 34 Median filters A Median Filter operates over a window by selecting the median intensity in the window. What advantage does a median filter have over a mean filter? Is a median filter a kind of convolution?
  • 35. 35 Matlab Lab 2 – Median Filter
  • 36. 36 Comparison: salt and pepper noise
  • 37. 37 Median Filter Example %%Median imrgb = imread('peppers.png'); imgray = im2double(rgb2gray(imrgb)); imnoise = imnoise(imgray,'salt & pepper',0.1); imshow(imnoise) figure imshow(medfilt2(imnoise))
  • 38. 38 Basic Filters summary Linear filtering is sum of dot product at each position Can smooth, sharpen, translate (among many other uses) Be aware of details for filter size, extrapolation, cropping 111 111 111

Editor's Notes

  1. e^0 = 1, e^-1 = .37, e^-2 = .14, etc.
  2. In matlab, conv2 does convolution, filter2 does correlation. Imfilter does either if specified, correlation by default (‘conv’, ‘corr’ option)
  3. Linear vs. quadratic in mask size
  4. f is image
  5. Better at salt’n’pepper noise Not convolution: try a region with 1’s and a 2, and then 1’s and a 3