0% found this document useful (0 votes)
4 views30 pages

Introduction

Image processing involves converting images into digital form to enhance them or extract useful information. It encompasses various operations, including visualization, image sharpening, retrieval, measurement, and recognition. The field faces challenges such as information loss during 3D to 2D conversion, noise, and ambiguity in visual data.
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
4 views30 pages

Introduction

Image processing involves converting images into digital form to enhance them or extract useful information. It encompasses various operations, including visualization, image sharpening, retrieval, measurement, and recognition. The field faces challenges such as information loss during 3D to 2D conversion, noise, and ambiguity in visual data.
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

INTRODUCTION

 
What  is  Image  Processing?  
• To  convert  an  image  into  digital  form  and  perform  
some  opera:ons  on  it,  in  order  to  get  an  enhanced  
image  or  to  extract  some  useful  informa:on  from  it.  
What  is  Image  Processing?  
• Computer  manipula:on  of  pictures,  or  images,  that  
have  been  converted  into  numeric  form.    
• Typical  opera:ons  include:    
o low-­‐level:  inputs  and  outputs  are  images  
o mid-­‐level:  outputs  are  acributes  extracted  from  input  
images  
o high-­‐level:  an  ensemble  of  recogni:on  of  individual  objects  
The  purpose  of  Image  Processing  

• Visualiza:on  -­‐  Observe  the  objects  that  are  not  


visible.  
• Image  sharpening  and  restora:on  -­‐  To  create  a  
becer  image.  
• Image  retrieval  -­‐  Seek  for  the  image  of  interest.  
• Measurement  of  pacern  –  Measures  various  objects  
in  an  image.  
• Image  Recogni:on  –  Dis:nguish  the  objects  in  an  
image.  
A  hard  task  
Vision  is  an  amazing  feat  of  natural  intelligence.  
 

• Visual  cortex  occupies  about  


50%  of  a  Macaque’s  brain  
 
• More  human  brain  devoted  to  
vision  than  anything  else  
Human  Vision  
• Recognising  objects  and  people  
• Locate  objects  in  space  
• Track  objects  in  mo:on  
• Recognise  ac:ons  
• But  humans  can  tell  a  lot  about  a  scene  from  a  licle  
informa:on…  
 
 

Source:  “80  million  :ny  images”  by  Torralba,  et  al.  


Human  Vision  
– We  see  effortlessly  and  immediately.  
– We  perceive  the  visual  world  as  external  to  ourselves,  but  it  is  
a  reconstruc:on  within  our  brains.  
– We  regard  how  we  see  as  reflec:ng  the  world  “as  it  is;”  BUT…  
human  vision  is…  
 
• subject  to  illusions  
• quan:ta:vely  imprecise  
• limited  to  a  narrow  range  of  frequencies  of  radia:on  
• passive  
Illusion  Examples  
Photometry  Overview  
Light  source  emits  
photon  
Sensor  Response  
IlluminaJon   Some  reaches  the  
eye/camera  and  
is  measured.  

Photons  travel  in  a  


straight  line  

They  hit  an  object.  Some  


gets  absorbed  while  some  is  
reflected  in  a  new  direc:on  
Surface  ReflecJon  
Fundamental  Challenges  
• 3D  to  2D  implies  informa:on  loss  

graphics  

vision  

• Noise  is  introduced  


• Ambiguity  –  different  things  may  look  similar  
Fundamental  Challenges  
• Appearance  changes  with  viewpoints  

• Geometric  changes:  surface  


slant  depends  on  viewpoint  

 
• Photometric  changes:  surface
brightness  and  colour  depend  on
viewpoint
• Occlusions:  what  is  hidden  depends  on
viewpoint
Fundamental  Challenges  
• We  are  trying  to  infer  things  about  the  world  from  an  array  of  
numbers  
Why  study  Image  Processing?  
• Images  and  videos  are  everywhere  

• And  they  are  diverse  


Why  Study  Image  Processing?  
• Fast-­‐growing  collec:on  of  useful  applica:ons  

Safety   Surveillance   Medical  

…  etc.  

Entertainment   Environment  
ApplicaJons  –  Medical  DiagnosJcs    
ApplicaJon  –  Face  DetecJon  and  RecogniJon    
• Many  new  digital  cameras  now  can  detect  faces  

• Sony’s  Smile  Shucer  Mode  


ApplicaJon  –  Image  based  Biometrics  

“How the Afghan Girl was


Identified?”
Image  Processing  System  
input  device  

AcquisiJon  
image  
Pre-­‐
processing  
image  

Problem   SegmentaJon   SoluJon  

image  regions  
DescripJon  and  
feature  extracJon  
RepresentaJon  
features  
PaWern  
learning  models  
ClassificaJon  
output  
Image  Processing  System  
input  device  

AcquisiJon  
image  
Pre-­‐
processing  
Post-­‐code  RecogniJon  
image   LN6  38PF  

Problem   SegmentaJon   SoluJon  

image  regions  
DescripJon  and  
feature  extracJon  
RepresentaJon  
11  black  pixels   features  
PaWern  
learning  models  
ClassificaJon  
three!   output  

 
Image  AcquisiJon  
Input:  Light,  Output:  Digital  Image  

Courtesy  of  [Link]  


Image  AcquisiJon  
Image  source  
• Light  -­‐  sensi:ve  camera,  scanner,  web-­‐cam,  etc.  
o usually  CCD  or  CMOS  sensor  
• Range  sensors,  tomography  devices,  radar,  
ultra-­‐sonic  cameras,  etc.  
• Output  of  most  sensors  is  a  con:nuous  value  
• Need  to  convert  it  into  digital  form  
 
Sampling  and  quan:sa:on  
• Digi:sing  the  coordinate  values  is  called  
sampling  
• Digi:sing  the  amplitude  values  is  called  
quan:sa:on  
 
Digital  Image  
The  image  data  contains  informa:on  of  the  gray  level  at  each  
posi:on  in  the  image.  

A  digital  image  of  a  rat.     The  magnifica:on  of  the  rat’s  nose.  
Digital  Image  
A  digital  image  can  be  thought  of  as  a  matrix  (2-­‐d  array,  raster  
map)  of  intensity  values  
  94 100 104 119 125 136 143 153 157 158

  103 104 106 98 103 119 141 155 159 160


109 136 136 123 95 78 117 149 155 160
  110 130 144 149 129 78 97 151 161 158

 
109 137 178 167 119 78 101 185 188 161
100 143 167 134 87 85 134 216 209 172

  104 123 166 161 155 160 205 229 218 181
125 131 172 179 180 208 238 237 228 200
  131 148 172 175 188 228 239 238 228 206

  161 169 162 163 193 228 230 237 220 199

The  magnifica:on  of  the  rat’s  nose.   Intensity  values  of  the  rat’s  nose.  
 
(common  to  use  one  byte  per  value:  0  =  black,  255  =  white)  
 
Pre-­‐processing  
Input:  Image,  Output:  Image  
Aim:  an  improvement  of  the  image  data  that  
suppresses  unwanted  distor:on  or  enhance  
some  image  features  important  for  further  
"Lena"  with  noise  
processing.  
 
Image  enhancement  
• adjust  contrast,  remove  noise,  sharpen  details  
Format  conversion   median  filtering  
• e.g.  from  RGB  to  HSV  colour  space  
Filtering  
• e.g.  remove  small  objects,  fill  in  holes  

edge  detec:on  
SegmentaJon  
Input:  Image,  Output:  Image  +  Regions  
Segment  (poten:ally)  relevant  image  regions  
• e.g.  foreground/background  detec:on  
• e.g.  skin  colour  filtering  
Based  on  1  or  2  basic  proper:es  of  intensity  values:  discon:nuity  
&  similarity  
• Such  as  thresholding,  region  growing,  region  spliwng  &  merging  
ROI  

thresholded
RepresentaJon  &  DescripJon  
Input:  Image  +  Regions,  Output:  Features  
 
Representa:on:  two  choices  
• In  terms  of  its  external  characteris:cs  (boundary)  
• In  terms  of  its  internal  characteris:cs  (pixels  comprising  the  region)  
 
Descrip:on:  based  on  the  representa:on  
• size,  shape  
• colour,  texture  
• manual  and  automa:c  extrac:on  

Examples  
• area  (all  black  pixels)  
• height  and  width  
16  black  pixels   31  black  pixels  
PaWern  ClassificaJon  
Input:  Features,  Output:  Class/Category  
 
Requires  learning  of  what  a  typical  class  ‘looks’  like  
(generalisa:on,  model)  –  based  (usually)  on  the  extracted  
features  
• object  recogni:on:  which  object  is  that?  
 
Classifier  
• training  –  learning  of  the  model  
• tes:ng  –  comparing  new  examples  to  the  learned  model  
PaWern  ClassificaJon  Example  
Task:  is  it  class  a  or  class  b?  
Training  
• A  data  set,  i.e.  a  large  number  of  example  images  (both  ‘a’  and  ‘b’)  
already  classified  by  a  human    
• Extract  features  for  each  example  
xn  =  black  pixel  count  

x1  =  black  pixel  count  


• Construct  a  histogram  for  each  class  –  count  the  number  of  examples  
with  a  specific  feature  value  

a  
frequency  

b  

feature  value  x  
PaWern  ClassificaJon  Example  
Task:  is  it  class  a  or  class  b?  
Training  (cont.)  
• Find  value  T  (discriminant)  that  will  best  separate  two  classes  (e.g.  
minimise  the  number  of  misclassifica:ons)  

a  
frequency  

b  

 
feature  value  x  
Tes:ng   T  
• extract  feature  x  from  a  given/new  example  
• if  x  <  T  then  the  new  example  is  a,  else  b  
input  device  

AcquisiJon  
image  
Pre-­‐
processing  
image  

SegmentaJon  

image  regions  
DescripJon  and  
RepresentaJon  
features  
PaWern  
ClassificaJon  
output  

You might also like