IMAGE PROCESSING FOR EVERYONE

George C Panayi, Alan C Bovik and Umesh Rajashekar

Laboratory for Vision Systems,

Department of Electrical and Computer Engineering
The University of Texas at Austin, Austin, TX 78712-1084 USA
{panayi, bovik, umesh}@ece.utexas.edu

ABSTRACT Engineering curriculum, EE 371R attracts many

undergraduate and graduate students from
The techniques of digital image processing have various departments, such as geology,
found a myriad of applications in diverse fields psychology, astronomy, computer science (to
of scientific, commerical, and technical name a few) as well as local industry. Image
endeavor. Image processing education therefore acquisition, image processing theory, and
needs to cater to a wide spectrum of people practical applications are introduced from an
coming from different educational backgrounds. operational perspective with some exposure to
In this paper, we describe tools and techniques theory. To encourage a web-based educational
that facilitate a gentle introduction to image system, all course material and visualization
processing. We present novel LabVIEW-based tools are made available to the students for
image processing demonstrations that, when downloading so that they can experiment with
supplemented with web-based class lectures, the tools at their leisure. MATLAB based
illustrate the power and beauty of image assignments help reinforce the concepts. The
processing algorithms. "Best Project Award" for the best DIP class
project (with a $100 first prize) motivates
1 Introduction students to investigate and apply the concepts
learned in the course to their respective fields.
Digital Image Processing (DIP) is a Beginning this Fall, students will also be
multidisciplinary science that borrows principles provided camcorders and webcams to develop
from diverse fields such as optics, surface their projects. Introductory material covered in
physics, visual psychophysics, computer science the course includes binary image processing,
and mathematics. The many applications of image analysis, and image enhancement, while
image processing include: astronomy, ultrasonic the more advanced material covers Hough
imaging, remote sensing, video communications transforms, edge detection and video
and microscopy, among innumerable others. processing. Since visualization is invaluable to
In this paper, we discuss teaching interpret the concepts, lecturing is accompanied
visualization tools developed for the EE 371R - with "in-class" LabVIEW demonstrations to
Digital Image and Video Processing course illustrate the concepts being discussed.
offered every Fall semester at the University of Section 2 of this paper gives a brief
Texas at Austin. The aim of the course is to overview of LabVIEW and its basic image
make DIP accessible to an audience of fairly processing functions, such as those embodied in
mixed backgrounds, using numerous visual the IMAQ Vision modules. In Section 3, we
examples to supplement the theory with explain some of the visualization modules we
reasonable mathematical simplicity. Though have created and we conclude in Section 4.
designed for an undergraduate Electrical

1
 2 LabVIEW applications, which run at compiled execution
speeds. Another advantage of LabVIEW is the
LabVIEW (Laboratory Virtual Instru- fact that it includes built in applications, such as
ment Engineering Workbench) is a graphical the IMAQ Vision for image processing. IMAQ
programming language used as a powerful and Vision includes more than 400 imaging
flexible instrumentation and analysis software functions and interactive imaging windows and
system in industry and academia. LabVIEW utilities for displaying and building imaging
uses a graphical programming language - G to systems. IMAQ Vision gave us the opportunity
create programs called Virtual Instruments or VI to create examples for all the important
(pronounced vee-eye) in a pictorial form called a functions in image processing, and use them for
block diagram, eliminating a lot of the educational purposes. [3] provides information
syntactical details of other programming on IMAQ Vision.
languages like C and MATLAB that use a text
based programming approach. LabVIEW also 3 Visualization modules
includes many tools for data acquisition,
analysis and display of results. The analysis The front panel provides an excellent
library contains a multitude of functions in intuitive graphical user interface (GUI) to vary
signal generation, signal processing, filtering various parameters in the algorithm. These
and statistics. LabVIEW is available for all the GUIs resemble the controls on many
major platforms and is easily portable across instruments and provide a user-friendly
platforms. Each VI contains 3 parts: interface (hence the name Virtual Instrument).
1) The front panel contains the user interface We have developed a wide range of VI’s
like knobs, push buttons, graphs and many other that can be used in conjunction with class
controls (inputs) and indicators (outputs). Inputs lectures. In this section, we describe few of the
can be fed using the mouse or the keyboard. Fig LabVIEW VIs that we developed. The reader
1a shows a typical front panel. can download other VI’s from:
2) The block diagram shown in Fig 1b is the http://pineapple.ece.utexas.edu/class/ee371r/Mo
VI’s source code constructed in G and is the dules/demos.htm. [4] provides detailed
actual executable program. The block diagram information on how the VI s were created.
has other lower-level VIs and built in functions.
The blocks can be connected using wires to 3.1 Analog to digital conversion
indicate the dataflow. Front panel objects have
corresponding terminals on the block diagram to Sampling and quantization help to
allow dataflow from the user to the program and transform the continuous domain image into a
back to the user. digital format. The effects of sampling and
3) Sub-VIs are analogous to subroutines in quantization can be visualized effectively with
conventional programming languages. the following two VI’s. This is a basic concept
[1] provides an introduction to that must be understood by any practitioner in
LabVIEW. [2] provides more information on any field utilizing digital images.
LabVIEW. In some applications such as image
processing, execution speed is critical. 3.1.1 Quantization
LabVIEW is the only graphical programming The Quantization VI (front panel in Fig
system with a compiler that generates optimized 2a) demonstrates the effects that different
code with execution speeds comparable to quantization levels have on images. This VI
compiled C programs. Thus, LabVIEW has the reads in an 8-bits/pixel image and creates an
ability to create stand-alone executable output image whose number of bit levels is

2
specified by the input parameter Number of bits. Dilation, Erosion, Open, Close, Open-Clos, and
The user has the options to create images that Clos-Open. The user has the option to test all
have 1, 2, 4, or 8 bit levels. Fig 2b shows the the above morphological operations on an
effects of quantization. image, and also to modify the type and size of
the structuring element. The Morphology VI
3.1.2 Sampling allows the following structuring element types:
To demonstrate the effects that different Row, Column, Square, Cross, and X-Shape.
sampling rates have on images, we created the
Sampling VI whose front panel shown in Fig 3.3 Histogram and point operations (Grayscale)
3a. While there is a mathematical theory of
sampling, the intuitive aspects of sampling can We developed VIs that perform linear
be understood by a diversity of users by visual (offset, scaling and full-scale contrast stretch)
observation of its effects. This VI reads in an and non-linear (logarithmic range compression)
image and sub-samples it to the user-specified image point operations. The VI for a linear point
size. The user has the option to sub-sample the operation is shown in Fig 4a. The user has the
image to the sizes of 256*256, 128*128, 64*64, option to perform the offset and scaling
and 32*32, using our VI. Fig 3b shows the sub- operations or to perform a full scale contrast
sampling of the input image to the specified stretch operation on an input image. The
size. The sub-sampled image is scaled to the histograms of the input image and the image
size of the input image, by duplicating columns after the linear point operation are also
and rows of the sub-sampled image. In this displayed in the front panel in the Histogram
image, we can see the effects that sampling has and New Histogram Waveform Graph
on images. The effects of sampling become indicators. Effects of linear point operations on
more pronounced when the sampling rate is an image are shown in Fig 4b.
decreased. Aliasing effects can be demonstrated
when severe undersampling occurs. 3.4 Image analysis (Frequency interpretations)

3.2 Binary Image processing 3.4.1 Discrete Fourier Transform (DFT)

The Fourier transform is fundamental to
Binary images have only two possible "gray image filtering and spectral theory, yet is a
levels" and can be represented using only one difficult concept for many users to correctly
bit per pixel. Besides developing VIs for grasp. We therefore begin by teaching the
thresholding grayscale images to binary, we also concept of digital frequency and digital
demonstrate the effects of binary morphology. sinusoidal gratings (2-D sine waves) as the basis
Morphological operations are defined by functions for the DFT. We have also constructed
moving a structuring element over the image to the FFT (Fast Fourier Transform) demonstration
be modified, in such a way that it is centered VI that calculates and displays the magnitude
over every image pixel at some point. When the and the phase of the DFT for gray level images.
structuring element is centered over a region of The front panel of this VI is shown in Fig 5a
the image, a logical operation is performed on Usually, the DFT is displayed with its centered
the pixels covered by the structuring element, coordinates (u, v) = (0, 0) at the center of the
yielding a binary output. We created the image. This way, the lower frequency
Morphology VI for demonstrating the effects of information is clustered together near the origin
various morphological operations on binary at the centered of the display. The Center/
images. In this VI, we implemented the seven Uncenter input parameter of the front panel
following morphological operations: Median, specifies whether the DFT will be displayed

3
with its low frequencies clustered together at the The other VI s developed includes linear and
center of the image or not. It is also the case non-linear filtering, image compression schemes
that the low-amplitude frequencies in the and a large number of edge detectors. We have
magnitude of the DFT will be hard to see, thus it presented the simplest ones here for the purpose
is best to logarithmically compress the DFT of illustration.
magnitude prior to display. An option is given
again for displaying the log compressed or 4 Conclusions
uncompressed magnitude of the DFT. This is
specified by the Compr/Uncompr input In this paper, we gave an overview of
parameter of the front panel. Fig 5b shows an Digital Image Processing education at UT-
image and the magnitude and phase of the DFT. Austin. We describe the use of powerful
visualization tools developed using LabVIEW
3.4.2 Directional DFTs for image processing. An overview of
When the DFT of an image is brighter LabVIEW and a few of the demonstrations
along a specific orientation, the image contains developed were provided.
highly oriented components in that direction.
Suppose that we define several oriented zero- 5 References
one images. Masking the DFT with these
images will produce IDFT images with only 1. Lisa K. Wells and Jeffrey Travis, "LabVIEW for
highly-oriented frequencies remaining. To Everyone, Graphical Programming made even
easier," Prentice Hall, 1997.
demonstrate this effect, we implemented the 2. ___,"Labview User Manual," National Instruments,
DFT Direction VI. The input parameters, Theta 1996.
1 and Theta 2, to this VI, are used as the starting 3. ___, "BridgeVIEW and LabVIEW IMAQ Vision for G
and final angles (in degrees) respectively, of the reference manual," National Instruments, 1996.
black region in the mask images as it can be 4. George C Panayi, "Implementation of Digital Image
Processing functions using LabVIEW," Master’s
seen from the “Circle” illustration in the front thesis, UT-Austin 1999.
panel in Fig 6a. Results of DFT masking are
shown in Fig 6b.

Fig 1a: Typical front panel

Fig 1b: Block diagram for front panel

4
Fig 2a: Front panel for Quantization Fig 2b: Original (left), Quantized(right)

Fig 3a: Front panel for sampling

Fig 3b: Original(left), sampled(middle), interpolated (right)

Fig 4a: Front panel for linear point operations

Fig 4b: Original(left), Offset and Scaling(middle), Full Scale stretch(right)

5
 Fig 5a: Front panel for DFT

Fig 5b: Original(left), DFT Magnitude(middle), DFT Phase(right)

Fig 6a: Front Panel for directional DFT

Fig 6b: Original (left), Result of directional DFT(middle), regions selected(Right)