HST.582J / 6.555J / 16.456J Biomedical Signal and Image Processing

MIT OpenCourseWare
http://ocw.mit.edu
HST.582J / 6.555J / 16.456J Biomedical Signal and Image Processing

Spring 2007
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.
Harvard-MIT Division of Health Sciences and Technology

HST.582J: Biomedical Signal and Image Processing, Spring 2007
Course Director: Dr. Julie Greenberg
HST-582J/6.555J/16.456J-Biomedical Signal and Image Processing-Spring 2007
Problem Set X
QUIZ 1 will take place on Tuesday, March 20, from 9:30-11 am in 56-154 (usual time and
place).
The quiz will cover material presented in lecture through Thursday, March 8.
The quiz will be closed book. One 8 12 11 inch sheet of notes (both sides) will be allowed.
Coverage of topics on the quiz will be somewhat representative of the amount of time spent
on each topic in lectures, labs and problem sets. You are not responsible for material in the
course notes that was not covered elsewhere in the course. Please see the annotated Outline
of Course Notes for listings of specific topics.
This ungraded problem set will help you prepare for the quiz in two ways. First, it will give
you a chance to think about solving problems in imaging processing, which was not covered
in any previous problem set. Second, it illustrates the type of questions asked on previous
years quizzes.
This problem set will not be collected or graded. Solutions will be posted on the course
website by Thursday, March 15.
Cite as: Julie Greenberg, Bertrand Delgutte, William (Sandy) Wells, John Fisher, and Gari Clifford. Course materials
for HST.582J / 6.555J / 16.456J, Biomedical Signal and Image Processing, Spring 2007. MIT OpenCourseWare
(http://ocw.mit.edu), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Problem 1
Consider the following small image for parts a) and b):
1
1
1
1
1
1
1
2
2
2
2
2
2
2
3 4
3 4
3 4
3 94
3 4
3 4
3 4
5
5
5
5
5
5
5
6
6
6
6
6
6
6
7
7
7
7
7
7
7
a) Calculate the image that results from convolution with the following kernel:
1 1 1
1 1 1
1 1 1
Dont worry about boundary values just provide the result for the central 5 by 5 part of
the result.
b) Calculate the image that results from applying a 3 3 median filter to the image described
above.
Again, dont worry about boundary values just provide the result for the central 5 by 5
part of the result.
Problem 2
The gradient of a two-dimensional function is a vector-valued derivative. The x component of the
(x,y)
gradient is defined by fx
.
Consider the following three discrete approximations to the x component of the gradient of an
image f . For each case, determine a convolution kernel that produces an image containing the
given approximation of the x component of the gradient.
a) f (n1 + 1, n2 ) f (n1 , n2 )
b) f (n1 , n2 ) f (n1 1, n2 )
c) .5 [f (n1 + 1, n2 ) f (n1 1, n2 )]
Problem 3
a) Consider the two-dimensional boxcar function defined to have unit amplitude when
|t1 | < T and |t2 | < T and zero otherwise. If this function is viewed as the impulse response
of a filter, how would you characterize the filter: high-pass, low-pass, or band-pass?
b) Next, examine the discrete analogue of the above function, which is defined to have unit
value when |n1 | < N and |n2 | < N. Now consider this function as a kernel for convolution
with some discrete 2D signal. In a brute-force convolution by this kernel, approximately
how many arithmetic operations are required per-pixel of the result?
c) The above discrete kernel is an example of a separable kernel. When taking advantage
of this separability in a convolution, approximately how many arithmetic operations are
required per-pixel of the result?
d) Next consider implementing the above convolution using FFT methods. Once again, approximately how many arithmetic operations are required per-pixel of the result, on a signal
with size M by M ?
e) Can you think of an implementation of the above 2D FIR filter, not using transforms, that
computes the result with approximately four arithmetic operations per-pixel of the result?
[Hint: exploit separabilty and use two passes of filtering, one with a horizontal kernel
and one with a vertical kernel. Focussing on the filtering by the horizontal kernel, and
assuming that you have just computed the result for some pixel, can you think of a cheap
way to compute the result for the neighboring pixel on the right?]
The following problems are taken from last years quiz. Please note
that the figures on pages 12 and 15 do not photocopy well, so you
may wish to view them online.
Question 1 (50%) This question has eight parts, but please dont panic; many of them can be
answered independently. Consider the continuous-time signal

sin(20t)
for t0 < t < t0
20t
x(t) =
0
otherwise
Figure 1 shows three functions
x(t) for t0 = 0.5
x[n], the discrete-time signal obtained by sampling x(t) at F s = 100 Hz
X(f ), the DTFT of x[n]
x(t)
x[n]
0.5
0.5
0
0.4
0.2
0.2
0.4
0.6
60
40
time (sec)
20
20
40
60
n
X(f)
A
0
C
B 0 B
frequency (f)
Figure 1:
(a) Determine the values of A, B, and C for X(f ), the DTFT of x[n], in Figure 1.
A=
B=
C=
(b) Is there a value of t0 that makes X(f ) a perfect rectangle? If so, state the value of t0 .
YES / NO
circle one
If yes, t0 =
(c) Is there a value of Fs that makes X(f ) a perfect rectangle? If so, state the value of Fs .
YES / NO
circle one
If yes, Fs =
Now consider a recently discovered alien species that has an ECG signal such that a single heartbeat
equals x(t) in Figure 1. A normal ECG reading for this alien, y(t), consists of a sequence of these
individual heartbeats defined by
y(t) = x(t) p1 (t) where p1 (t) =
(t r).
r=
An abnormal ECG reading, z(t), consists of a sequences of these individual heartbeats alternating
with segments of zero signal, defined by
z(t) = x(t) p2 (t) where p2 (t) =
(t 2r).
r =
y(t) normal
1
0.5
0
0.5
10
10
z(t) abnormal
1
0.5
0
0.5
time (sec)
Figure 2:
(d) z(t) is the result of convolving x(t) with an impulse train with a period of T = 2 sec. Circle
the appropriate choice on each line to make the following sentence true.
Z(F ), the CTFT of z(t), will be the result of adding / multiplying / convolving
X(F ) with a periodic boxcar / sinc function / impulse train
with period 0.25 Hz / 0.5 Hz / 1 Hz / 2 Hz / 4 Hz.
Figure 3 shows the results of frequency analyses using the Matlab function pwelch:
pwelch(signal,rectwin(WINLEN),0,NFFT,Fs,twosided).
Note that all analyses used a rectangular window with no overlap.
(e) Which plot(s) in Figure 3 show(s) a frequency analysis of y(t)?
A / B/ C / D / E
circle all that apply
Justification:
(f) Which plot(s) in Figure 3 show(s) a frequency analysis of z(t)?

A/B/C/D/E
circle all that apply
Justification:
(g) Which plot in Figure 3 was generated using used the shortest window length?
A/B/C/D/E
circle one
Justification:
(h) All of the analyses in Figure 3 used a rectangular window. How will the plots change if a
Hamming window of the same length is used?
spikes will be wider /narrower / unchanged
circle one
average magnitude of regions between spikes will increase / decrease / stay same
spikes will be closer together /farther apart / spacing unchanged
circle one
circle one
Question 2 (15%)
Consider the signal
x(t) = P + Q sin(2F t) + Rn(t)
where
P, Q, and R are constants,
F = F1 for t < t1 sec,
F = F2 for t > t1 sec , and
n(t) is highpass noise above FH .
x(t) is lowpass filtered with an analog filter with cutoff frequency Fc and sampled at Fs = 10 kHz
to produce x[n].
(a) Figure 4 shows two spectrograms of x[n]. Estimate the values of the following parameters.
Fc
F1
F2
FH
(b) What spectrogram analysis parameter was changed to create the two different plots in Figure
4?
In Questions 3 and 4, lower case letters refer to the sample space description while upper case
letters refer to the frequency space description, for example X[k1 , k2 ] is the DFT of x[n1 , n2 ]:
x[n1 , n2 ] X[k1 , k2 ]
x[n] X[k]
It is not necessary to compute the DFT of any function in order to answer the questions below.
Question 3 (20%)
Given the following 1-D filters
h1 [n]
h2 [n]
g1 [n]
g2 [n]
and separable 2-D filters defined as follows

h[n1 , n2 ] = h1 [n1 ]h2 [n2 ]
h[n1 , n2 ] = g1 [n1 ]g2 [n2 ]
answer parts (a) and (b).
(a) Which of the system block diagrams shown in Figure 5 have a frequency response equivalent
to:
H[k1 , k2 ]G[k1 , k2 ]
(CIRCLE ALL THAT APPLY)
A
NONE
(b)
The system block diagram shown in Figure 6 has a frequency response of:
H[k1 , k2 ] G[k1 , k2 ]
Find expressions for separable filters a[n1 , n2 ] and b[n1 , n2 ] in terms of h1 [n1 ], g1 [n1 ],
g2 [n1 ].
h2 [n2 ],
a[n1 , n2 ] =
b[n1 , n2 ] =
Question 4 (15%)
Consider the 4848 input image in Figure 7, with values coded from 1 for black to +1 for white,
as shown. This image contains three values: 1 in the darkest regions, 0 around the border, and
1 in the three light squares. Identify which of the output images in Figure 8 result from applying
h1 [n1 , n2 ] and h2 [n1 , n2 ] to the original input image, where h 1 [n1 , n2 ] and h2 [n1 , n2 ] are defined
as:
1 1 1 1 1
0 13 0
25
25
25
25
25
1 1 1 1 1
0 1 0

25 25 25 25 25
1 1 1 1 1
h2 [n1 , n2 ] = 25
h1 [n1 , n2 ] = 0
0 0
25
25
25
n2
1 25
1
1
1
1
1 n2
25 25 25 25 25
0
0
6
0
n1
h1 [n1 , n2 ]:
1
3
1
25
1
25
n1
1
25
(circle one)
(circle one)
1
25
1
25
Justification:
h2 [n1 , n2 ]:
Justification:
10
A
50
40
30
20
10
10
20
30
40
50
40
30
20
10
10
20
30
40
50
40
30
20
10
10
20
30
40
50
40
30
20
10
10
20
30
40
50
40
30
20
10
10
20
30
40
50
B
50
Figure 3:
11
C
50
D
50
E
50
frequency (Hz)
5000
4500
frequency (Hz)
4000
3500
3000
2500
2000
1500
1000
500
0
0.02
0.04
0.06
0.08
0.1
0.14
0.16
0.18
0.2
0.12
0.14
0.16
0.18
0.2
time (sec)
Figure 4:
5000
4500
4000
frequency (Hz)
12
0.12
3500
3000
2500
2000
1500
1000
500
0
0.02
0.04
0.06
0.08
0.1
time (sec)
h1 [n1 ] [n2 ]
[n1 ] g2 [n2 ]
g1 [n1 ][n2 ]
[n1 ]h2 [n2 ]
h1 [n1 ]h2 [n2 ]
h1 [n1 ]g2 [n2 ]
g1 [n1 ]h2 [n2 ]
g1 [n1 ]g2 [n2 ]
h1 [n1 ] [n2 ]
[n1 ] g2 [n2 ]
g1 [n1 ][n2 ]
[n1 ]h2 [n2 ]
h1 [n1 ]g1 [n1 ]h2 [n2 ]g2 [n2 ]
Figure 5: System Block Diagram for Question 3(a).
13
(h1 [n1 ] + g1 [n1 ])(h2 [n2 ] g2 [n2 ])
a(n1 , n2 )
b(n1 , n2 )
Figure 6: System Block Diagram for Question 3(b).
14

n2
n1
Figure 7: Original image
(A)
(B)
(C)
(D)
Figure 8: Processed images
15

HST.582J / 6.555J / 16.456J Biomedical Signal and Image Processing

Uploaded by

Copyright:

Available Formats

HST.582J / 6.555J / 16.456J Biomedical Signal and Image Processing

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

HST.582J / 6.555J / 16.456J Biomedical Signal and Image Processing

Uploaded by

Copyright:

Available Formats

MIT OpenCourseWare

HST.582J / 6.555J / 16.456J Biomedical Signal and Image Processing

Harvard-MIT Division of Health Sciences and Technology

HST-582J/6.555J/16.456J-Biomedical Signal and Image Processing-Spring 2007

circle all that apply

(f) Which plot(s) in Figure 3 show(s) a frequency analysis of z(t)?

circle all that apply

and separable 2-D filters defined as follows

[n1 ]h2 [n2 ]

h1 [n1 ]h2 [n2 ]

h1 [n1 ]g2 [n2 ]

g1 [n1 ]h2 [n2 ]

g1 [n1 ]g2 [n2 ]

[n1 ]h2 [n2 ]

h1 [n1 ]g1 [n1 ]h2 [n2 ]g2 [n2 ]

Figure 5: System Block Diagram for Question 3(a).

(h1 [n1 ] + g1 [n1 ])(h2 [n2 ] g2 [n2 ])

Figure 6: System Block Diagram for Question 3(b).

Figure 7: Original image

Figure 8: Processed images

You might also like