Computer Vision &

unclesamnorweiganΤεχνίτη Νοημοσύνη και Ρομποτική

18 Οκτ 2013 (πριν από 3 χρόνια και 7 μήνες)

82 εμφανίσεις

john@hulskamp.com.au

1

Computer Vision &
Image processing

John Hulskamp

E
-
mail: john@hulskamp.com.au

Consultant

john@hulskamp.com.au

2

Introduction


Introducing the new subject 31049
Computer Vision & Image
Processing



Guest Lecturer with previous
experience in teaching & research
in the area


john@hulskamp.com.au

3

Osteons


Compact bone is comprised of many units called osteons: they consist of a central
canal surrounded by closely packed concentric layers called lamallae.


This is an image of some osteons, specimen is from a 32yo male. Image is a
~2.8x3.5mm portion of a contact micro
-
radiograph taken from a 100micrometre
thick, un
-
embedded, hard
-
ground section. Black areas are voids, white = high
mineral density (maybe a couple of grains of carborundum can be seen).


Attribution: David Thomas, Dental Sciences, University of Melbourne


john@hulskamp.com.au

4

Near IR Imaging


Hot strip mill temperature measurement

john@hulskamp.com.au

5

Near IR Imaging (2)


Thermal Map


Attribution: C. Lampe: “A Multi
-
Processor temperature profiling system
for real time control in a hot strip rolling mill” M.Eng. Thesis RMIT 1995


john@hulskamp.com.au

6

Topics for Discussion
Today


Subject objectives


Fundamental Issues


Applications


Towards image understanding


CV & IP Tools


A Simple Beginning


Conclusions




john@hulskamp.com.au

7

Subject objectives


adequate background knowledge about
computer vision and image processing



practical knowledge and skills about
computer vision and image processing tools



necessary knowledge to design and
implement a prototype of a computer vision
application


john@hulskamp.com.au

8

Fundamental Issues


An imaging system


john@hulskamp.com.au

9

Fundamental Issues (2)


Resolution
-

the smallest feature size on your object that
the imaging system can distinguish


Field of view
-

the area of inspection that the camera
can acquire


Working distance
-

the distance from the front of the
camera lens to the object under inspection


Sensor size

-

the size of a sensor's active area


Depth of field
-

the maximum object depth that remains
in focus

john@hulskamp.com.au

10

Fundamental Issues (3)

Resolution


Resolution indicates the amount of object detail that the imaging system can
reproduce. You can determine the required resolution of your imaging system by
measuring in real
-
world units the size of the smallest feature you need to detect in the
image.


To make accurate measurements, a minimum of two pixels should represent the
smallest feature you want to detect in the digitized image. In the picture , the
narrowest vertical bar (w) should be at least two pixels wide in the image. This
information can assist to select the appropriate camera and lens for the imaging
application.

john@hulskamp.com.au

11

Fundamental Issues (4)


Sensor resolution

is the number of columns and rows of CCD pixels in the
camera sensor. To compute the sensor resolution, you need to know the field of
view (FOV).


The FOV is the area under inspection that the camera can acquire. The horizontal
and vertical dimensions of the inspection area determine the FOV. Make sure the
FOV encloses the object you want to inspect.


Once you know the FOV, you can use the following equation to determine your
required sensor resolution:


sensor resolution = (FOV/resolution) x 2

= (FOV/size of smallest feature) x 2


john@hulskamp.com.au

12

Fundamental Issues (5)

Determine the focal length of your lens.


A lens is primarily defined by its focal length. This picture illustrates the relationship
between the focal length of the lens, field of view, sensor size, and working
distance which is the distance from the front of the lens to the object under
inspection.


focal length = sensor size x working distance / FOV

john@hulskamp.com.au

13

Fundamental Issues (6)

Lighting issues

One of the most important aspects of setting up your imaging environment is proper
illumination. Images acquired under proper lighting conditions make your image
processing software development easier and overall processing time faster. One
objective of lighting is to separate the feature or part you want to inspect from the
surrounding background by as many gray levels as possible. Another goal is to control
the light in the scene. Set up your lighting devices so that changes in ambient
illumination
-
such as sunlight changing with the weather or time of day
-
do not
compromise image analysis and processing.

john@hulskamp.com.au

14

Fundamental Issues (7)


Backlighting
is another lighting technique that can help improve the
performance of your vision system. If you can solve your application by
looking at only the shape of the object, you may want to create a silhouette
of the object by placing the light source behind the object you are imaging.
By lighting the object from behind, you create sharp contrasts which make
finding edges and measuring distances fast and easy. This picture shows a
stamped metal part acquired in a setup using backlighting.

john@hulskamp.com.au

15

Fundamental Issues (8)

Perspective

Perspective errors occur when the camera axis is not perpendicular to the object under
inspection. The figures show both an ideal camera position (I.e. vertical) and a
camera imaging an object from an angle.

john@hulskamp.com.au

16

Fundamental Issues (9)

Perspective and Distortion Errors


Try to position your camera perpendicular to the object under inspection to
reduce perspective errors. Integration constraints may prevent you from
mounting the camera perpendicular to the scene. Under these constraints,
you can still take precise measurements by correcting the perspective
errors with spatial calibration techniques.


john@hulskamp.com.au

17

Examples of Vision
Applications


Attribution: National Instruments IMAQ Vision Product
Demonstration (
www.ni.com/vision
)




Examples:


Battery Clamp


Spark Plug Gap Measurement


Blister Pack Inspection


PCB Inspection

john@hulskamp.com.au

18

Examples of Vision
Applications (2)


Battery Clamp

john@hulskamp.com.au

19

Examples of Vision
Applications (3)


Spark Plug Gap Measurement

john@hulskamp.com.au

20

Examples of Vision
Applications (4)


Blister Pack Inspection:

Ensuring that blister packs contain the correct number
and type of pills before they reach pharmacies, ensuring the integrity of the product
and increase the yield of production by automating the inspection of blister pack
contents.









Acquire colour images of the blister packs. Use colour location to count the number of
green areas in the image. With colour location, you create a model or template that
represents the colours that you are searching. Then the machine vision application
searches for the model in each acquired image and calculates a score for each
match. The surface area of each pill in the pack must be at least 50% green to pass
inspection.


john@hulskamp.com.au

21

Examples of Vision
Applications (5)


Blister Pack Inspection (cont’d)

john@hulskamp.com.au

22

Examples of Vision
Applications (6)


PCB Inspection


To ensure that components are present and at the correct
orientation on a PCB.












john@hulskamp.com.au

23

Examples of Vision
Applications (7)


PCB Inspection

(cont’d):

Colour information simplifies a monochrome
problem by improving contrast or separation of the components from the background.
Colour pattern matching can distinguish objects from the background more efficiently
than grayscale pattern matching.


This example uses rotation
-
invariant pattern matching because it can detect the
components regardless of their orientations. You can use the orientation information
to determine the correct placement of orientation
-
sensitive components, such as
capacitors or diodes.



john@hulskamp.com.au

24

Towards image
understanding


Computer Vision: making useful decisions about real physical objects
and scenes based on sensed images



To get to that understanding we need to process images: hence
Image Processing is a step towards this understanding



Image Processing issues:


Image Statistics & Histograms


Image Enhancement


Image Restoration


Image Analysis: Edge Detection & Feature extraction


Representation & Description


Pattern Recognition

john@hulskamp.com.au

25

Histograms


The cumulative histogram of a grey
-
level image
f(w,h)

is a
function
H(k)

which provides the total number pixels (number
of occurrences) that have grey
-
level less than the value
k
.








Where
card

stands for the cardinality (i.e. number of pixels)
of a set.


The histogram of a grey
-
level image
f(w,h)

is a table
h(k)

which is the discrete difference of the cumulative histogram. It
provides the total number pixels (number of occurrences) that
have a specific grey
-
level value
k
.


Attribution: www.khoral.com

john@hulskamp.com.au

26


Histograms (2)



john@hulskamp.com.au

27

Histogram Equalisation



john@hulskamp.com.au

28

CV & IP Tools


NIH

Image

is a public domain image processing and analysis program
for the Macintosh. It was developed at the
Research Services Branch
(RSB)

of the National Institute of Mental Health (NIMH), part of the
National Institutes of Health (NIH)
. A free PC version of
Image
, called
Scion Image for Windows, is available from
Scion Corporation
.
Image

can acquire, display, edit, enhance, analyse and animate images. It
reads and writes TIFF, PICT, PICS and MacPaint files, providing
compatibility with many other applications, including programs for
scanning, processing, editing, publishing and analysing images. It
supports many standard image processing functions, including
contrast enhancement, density profiling, smoothing, sharpening, edge
detection, median filtering, and spatial convolution with user defined
kernels.


Image

can be used to measure area, mean, centroid, perimeter, etc. of
user defined regions of interest. It also performs automated particle
analysis and provides tools for measuring path lengths and angles.
Spatial calibration is supported to provide real world area and length
measurements. Density calibration can be done against radiation or
optical density standards using user specified units. Results can be
printed, exported to text files, or copied to the Clipboard.


john@hulskamp.com.au

29

CV & IP Tools (2)


Khoral Inc:
It started out as Khoros about 10 years ago, for UNIX
-
based X
-
systems, has become available for the Windows platform as
well. Costs money today!


Their online DIP course is very good:
http://www.khoral.com/contrib/contrib/dip2001/index.html

john@hulskamp.com.au

30

CV & IP Tools (3)


UTHSCSA ImageTool


UTHSCSA ImageTool (IT) is a free image processing and analysis
program for Microsoft Windows 9x, Windows ME or Windows NT. IT
can acquire, display, edit, analyse, process, compress, save and print
gray scale and colour images.IT can read and write over 22 common
file formats including BMP, PCX, TIF, GIF and JPEG. Image analysis
functions include dimensional (distance, angle, perimeter, area) and
gray scale measurements (point, line and area histogram with
statistics). ImageTool supports standard image processing functions
such as contrast manipulation, sharpening, smoothing, edge detection,
median filtering and spatial convolutions with user
-
defined convolution
masks.

ImageTool was designed with an open architecture that provides
extensibility via a variety of plug
-
ins. Support for image acquisition
using either Adobe Photoshop plug
-
ins or Twain scanners is built
-
in.
Custom analysis and processing plug
-
ins can be developed using the
software development kit (SDK) provided (with source code). This
approach makes it possible to solve almost any data acquisition or
analysis problem with IT.



john@hulskamp.com.au

31

CV & IP Tools (4)


Intel open source computer vision library
OpenCV



CVIPtools: is a UNIX/Win32
-
based software package developed
\

under the continuing direction of
Dr. Scott E Umbaugh
One of the
primary purposes of the CVIPtools development is to allow students,
faculty, and other researchers to explore the power of computer
processing of digital images.


CVIPtools is a
collection
of computer imaging tools providing services
to the users at three layers. At the bottom level are the
CVIPtools
libraries

(the application programming interface). Based on the
CVIPtools libraries are the cviptcl and cvipwish shells. The cviptcl shell
is an extension of
Tcl

with additional CVIP capabilities. With cviptcl,
the user can either use the command line for interactive image
processing, or write cviptcl shell scripts for batch processing. The
cvipwish shell is the extension of cviptcl with the added functionality for
building a graphical user interface (GUI) which allows even the casual
computer users to experiment with many of the sophisticated tools
available to computer imaging specialists without the need for any
knowledge of computer programming.




john@hulskamp.com.au

32

Topics in Course


Image formation.

Human vision. Computer vision.


Image representation.

Analog and digital images. Common image and
video formats. Image compression. JPEG and MPEG.


Image acquisition.

Image acquisition systems. Cameras and frame
grabbers.


Image processing.

Image enhancement. Convolution. Filtering. Edge
detection. Texture analysis. Labelling. Contour tracing. Image
morphology. Image segmentation.


Image analysis.

Feature extraction. Geometrical features. Hough
transform.


Image sequences.

Motion detection. Optical flow. Background
subtraction. Feature tracking.


Pattern recognition.

Object classification. Statistical, neural networks,
symbolic classifiers.


Computer Vision.

Model
-
based vision. Applications in
industrial
quality inspection, video surveillance, robotics, medicine,
multimedia
.



john@hulskamp.com.au

33

A Simple Beginning


Paradigm for an image processing
code:

Declarations


File In: Bring in image into Array I[r,c]


Processing on Array I[r,c] to produce

Array O[r,c]


File Out: Output Array O[r,c] as image



john@hulskamp.com.au

34

Conclusions


An overview of what is ahead for
you in this exciting field.


You might like to experiment with
the above simple code to attempt
to gain the histogram for an image,
and try out “histogram
equalisation”