Course No.

: EEM722 / PEE 316, Title: DIGITAL IMAGE PROCESSING

Class: B.Tech., Status of Course: Major Course, Approved since session: 2009-10

Total Credits:3, Total pds.(55 mts each)/week: 3(L:3+T:0+P:0+S:0)Min. pds./sem.:39

UNIT 1: IMAGE FUNDAMENTALS

Visual perception by human eye, Brightness Adaptation and Discrimination, Image Models,
Sampling and quantization, Colour models, File formats.

UNIT 2: IMAGE ENHANCEMENT IN SPATIAL DOMAIN

Point Operations: Image negative, contrast stretching, Brightness, Grey level slicing, Bit plane
slicing, Histogram Processing.
Spatial Operations: Smoothing filters, Median Filter, Sharpening filters, High boost filtering,
derivative filtering, Robert, Previtt, Sobel operators, Second order derivatives, Laplacian Mask.

UNIT 3: IMAGE ENHANCEMENT IN FREQUENCY DOMAIN

Review of Fourier Transforms, Discrete Fourier Transforms, 2-dimensional DFT, Low pass
(smoothing) filters, High pass filters.

UNIT 4: IMAGE SEGMENTATION

Point detection, line detection, edge detection, combined detection, Edge linking and boundary
detection-Hough transforms, Thresholding.

UNIT 5: IMAGE COMPRESION

Fundamentals: Coding redundancy, Inter-pixel redundancy, Psycho-visual redundancey, Fedility
criterion.
Compression Models: Source encoder and decoder, Channel encoder and decoder.
Lossless compression: Variable length coding, bit plane coding, lossless predictive coding.
Lossy compression: Lossy predictive coding, Transform coding.

SUGGESTED READINGS:
RC Gonzalez & RE Woods: DIGITAL IMAGE PROCESSING, Pearson Education
AK Jain: FUNDAMENTALS OF DIGITAL IMAGE PROCESSING, PHI
COMPUTER VISION HOME PAGE: Http://www.cs.cmu, edu/~cil/txtvision. Html
 Dayalbagh Educational Institute
BTech Final Year (EEM 722)
MTech (Engineering Systems) (PEE 316)
Digital Image Processing
Question Bank 2017 -18

Unit 1: Image Fundamentals

1. A common measure of transmission for digital data is the baud rate, defined as the number of bits
transmitted per second. Generally, transmission is accomplished in packets consisting of a start bit, a
byte (8 bits) of information, and a stop bit. Using this approach, answer the following.

(i) How much time would it take to transmit a 512x512 image with 256 gray levels at 300 baud?
(ii) What would the time be at 9600 baud?

2. An image f(x,y) is approximated by equally spaced samples arranged in the form of an array of the
size NxM. Each element of the array represents a gray level using a discrete quantity between (0 to L-1).
What is the storage requirement in terms of number of bits for the image?

3. What are the two discretization processes that are necessary to give the digital model of an image
defined as f(x, y)? What is the effect of reducing the spatial resolution and intensity resolution (bits per
pixel) on the quality of the image? Specify the kind of artifacts that can be observed in the two cases.

4. Explain how the D4 distance between two points p(x, y) and q(s, t) is equal to the shortest 4-path
between these points. Consider the following part of an image with intensity values of 0 or 1.
1001
1 0 1(q) 1(r)
1 1(p) 0 0
1111
(i) Find the D4 and D8 distances between p and r pixels in the above image.
(ii) For V={1} determine whether p and q are 4-connected or 8-connected.

5. An 8-bit pixel bi-level (two intensity values) image is given as follows. The region A has the intensity
value 120 and the region B has the intensity value of 60.
(i) Draw the intensity profile of the image at the section X-X.
(ii) Will the intensity profile as perceived by human visual system would be different along X-X? If so
draw it. If not so explain why?

6. You go from the blazing sunlight into a dark auditorium and you take a while to adjust to the new
surroundings and then you are able to see well. Explain the reason for this property of the Human Visual
System.

7. Where is C (Cyan) M (Magenta)Y (Yellow) color model used? How can you obtain CMY model from the
RGB model?

8. Suggest a color model that decouples the chrominance (color) component and luminance (intensity)
component. What is the advantage of such a color model?
9. The following image (with 8 bit pixels) has the background indicated by B and foreground indicated by
A. The intensity of the foreground is constant and fixed to 150. The background intensity is also constant
but has three cases with intensities as (i) 50 (ii) 0 (iii) 100. Will the perceived brightness/intensity be the
same in all cases? Justify your answer.

10. A part of a row of an image with 8 bit pixels is 223 220 160 144 128 Construct a 4 bit pixel
image for the same part.

11. Compare the roles of rods and cones in human vision system.

12. Distinguish between (a) Photopic and scotopic vision (b) Subjective brightness and actual brightness.

13. Explain the following terms in the context of human visual system (a) Brightness Adaptation (b)
Weber Ratio (c) Simultaneous Contrast.

14. Explain the following phenomena in images and state how they can be avoided (a) Sampling
checkboards (b) False contouring.

15. Give a high level description of the following systems using a block diagram (a) License plate
verification system for auotmobiles at an underground parking entrance (b) OCR for printed Roman text.

16. Give some applications of image processing in (a) military (b) medical domain.

Unit 2: Image Enhancement in Spatial Domain

1. What is meant by image enhancement? Why is it said that there is no one size fits all solution for
image enhancement?

2. Assume you are given an image that suffers from the following problems related to image quality.

a. The image does not have enough contrast. Most areas in the image appear to be too
bright.
b. The structures and boundaries in the image are blurred and thus it is hard to see the details
of objects in the image.
c. There are random sparse black spots (pepper noise) that seem to be caused by some
electronics noises.
You are asked to propose a system that use techniques you have learned in this class to improve the
overall image quality. Design a conceptual diagram for a quality enhancement system that addresses all
the problems mentioned above. Provide justifications for the use of each component and the specific
order you adopt in combining different components. Try to provide as much information as needed. For
example, if you use contrast stretching, specify the shape of the intensity mapping function. If you use
sharpening filters, specify the specific type of filter you will use. Also state your assumptions if any.
 3. A typical transformation used for contrast stretching is shown below. Why in general one keeps
the values of r1, r2, s1, and s2 as r1 r2 and s1 s2 ? For what values of r1, s1, r2, s2 one can obtain
a binary image?

4. An image has the gray level probability density function (PDF) as shown below. It is desired to
transform the gray levels of this image so that they will have the specified p(z) as shown below. Assume
continuous quantities and find the transformation in terms of r and z that will accomplish this.

5. Consider the 4x4 image as given below. What is the result of applying a linear smoothing filter on
this image. What is the result of applying a median filter? Show the computation and comment on
the result.
13 5 11 6
15 4 6 7
12 14 4 10
10 12 5 11

6. Specify the histogram of the image shown below in tabular form as well as graphical form.
13 5 11 6
15 4 6 7
12 14 4 10
10 12 5 11

7. Consider the following 8-level gray image of size 8x8. Show the original histogram of the image.
Compute the equalized histogram and display graphically.

0 1 1 0 0 1 1 0
0 1 2 3 3 2 1 0
0 4 5 3 3 5 4 0
0 1 2 7 7 2 1 0
0 1 2 6 6 2 1 0
0 4 5 3 3 5 4 0
0 1 2 3 3 2 1 0
0 1 1 0 0 1 1 0
8. Propose a set of gray-level slicing transformations capable of producing all the individual bit planes
of an 8-bit monochrome image.

9. What happens if you apply histogram equalization repeatedly on an image? Justify your answer.

10. For digital images it is observed that when histogram equalization is applied the result is not a flat
(uniform) histogram. What may be the reason?

11. While implementing the low pass filter in spatial domain, you are required to write a procedure of
moving the mask (say of size nxn) on the image and compute the response by multiplying the
image intensity value with the mask coefficient and summing them up. Suggest a scheme of
updating only the part of computation that changes from one location to another, let the
coefficients used for the mask be 1. What is the computational advantage using this scheme, if
any?

12. Histogram equalization is generally considered as a global operation for enhancement of images.
How can histogram equalization be performed for local enhancement? Suggest a computationally
efficient scheme of performing histogram equalization for local enhancement.

13. How does Median filter help in the removal of noise with dark and/or white spots? Can you use the
neighborhood average filter instead? How does the size of the mask affect in the computation for
the Median filter, how would you compare this computational aspect with the neighborhood
method?

14. Consider the following two images of equal size having pixel values of 0 (black) and 255 (white).
The divisions of the white and black parts are also of the same sizes in both the images.

a. Will the histogram of the two images be the same? Plot the histogram in each case.
b. A neighborhood-averaging mask of size 3x3 for smoothing is applied on the image. What
will happen to the histogram to the two images? Plot the histogram in the two cases.

15. You have 50 images from an experiment in biology, which are all noisy. These images are of the
same sample and are perfectly aligned. The noise is random and has got zero mean. Suggest a
scheme of obtaining an image with reduced noise and give your argument why would your scheme
work. Assume that images have the intensity range as 0-255.

16. (a) What effect would setting to zero the lower-order bit planes have on the histogram of an image
in general. (b) What would be the effect on the histogram if we set to zero the higher-order bit
planes instead?

17. Two images f(x,y) and g(x,y) have histograms h f and hg. Give the conditions in which you can
determine the histograms of

a. f(x,y) + g(x,y)
b. f(x,y) - g(x,y)
c. f(x,y) x g(x,y)
 18. How can image averaging be used for image enhancement? Give examples.

19. Consider the following image having gray values between levels 0 and 15. Stretch the gray values
uniformly and linearly within the entire range.

10 11 12 11 10
12 12 13 5 4
13 12 5 3 5
13 12 4 3 5
12 4 5 4 4

20. Give an example each of a (a) Spatial low pass filter (b) Order statistics filter (c) Sharpening filter
in spatial domain.

Unit 3: Image Enhancement in Frequency Domain

1. Give the formulae for the Fourier Transform and Inverse Fourier Transform in 1-d in the (i)
Continuous case (ii) Discrete Case.

2. Explain the following terms in the context of 1-d Fourier Transforms (i) Spectrum of the Fourier
Transform (ii) Phase angle of the Fourier Transform (iii) Power Spectrum of the Fourier Transform

3. Give the formulae for the Fourier Transform and Inverse Fourier Transform in 2-d in the (i)
Continuous case (ii) Discrete Case.

4. Explain the following terms in the context of 2-d Fourier Transforms (i) Spectrum of the Fourier
Transform (ii) Phase angle of the Fourier Transform (iii) Power Spectrum of the Fourier Transform

5. Describe the main steps in performing filtering in the frequency domain with the help of a suitable
example.

6. Consider the following 2x2 image and the transformation matrix A. Find the basis matrices.

1 2 1 1
U=[ ] and A = 1/2 [ ]
3 4 1 1

7. Show that one can construct the original image from its unitary transformation coefficients given
in Q6.

8. Find the Fourier Spectrum of the 1-d constant function as f(x) = C for 0 x N.

9. Derive the Fourier Transformation matric for a 4x4 image.

10. A function f(x) is sampled at 4 points x = 0.5, 0.75, 1.00, and 1.25 respectively and the sampled
values are f(x) = {2,3,4,4} respectively. Find the Fourier transform coefficients for the sampled
function.

11. A high pass filtered image can be obtained by subtracting a low pass filtered image from the original
image. Show that this is true both in spatial and frequency domains. For spatial domain you may
consider a mask of 3x3.
 12. How many additions and multiplications would be required for computing 2-D Fast Fourier
Transform (FFT) of an image of NxN? How do you compare them with normal Discrete Fourier
Transform (DFT)?

13. Roberts gradient operators use simple difference of pixel values for designing a sharpening filter
in spatial domain. Find the equivalent filter function in the frequency domain applied to an image
with its average values as favg(x,y). Compute the new average value of the image after applying
this filter.

14. (a) An image f(x,y) has its Fourier transform as F(u,v). The image f(x,y) is rotated by an angle
theta about the origin. What will happen to the Fourier transform of the rotated image? (b) How
can you compute the inverse Fourier transform knowing the forward Fourier transform?

15. What is the effect of changing the cut off frequency D 0 in an ideal low pass filter in the frequency
domain? Under what conditions the Butterworth low pass filter behaves like an ideal low pass filter
in the frequency domain?

16. What are the advantages and disadvantages of designing the filter functions in the frequency
domain?

17. Often it is desirable to shift the origin of the Fourier transform of an image f(x,y) to the center of
its corresponding NxN frequency square (i.e., to N/2 x N/2). How do you achieve this? What
happens to the magnitude of the Fourier transform?

Unit 4: Image Segmentation

1. How do you detect edge using the gradient operator? Mention some of the operators used in
practice.

2. The results obtained by a single pass through an image of some 2-D masks can also be achieved
by two passes using 1D masks. For example, a 3x3 smoothing masks can be obtained by first
passing through an image a horizontal mask [1 1 1] and then a vertical mask [1 1 1] T . Show that
the 2D mask for the Sobel operator can be obtained if two 1D masks are applied one row wise for
smoothing [1 2 1]and the other differencing mask [-1 0 1] column wise.

3. Why is Sobel operator preferred over other gradient operator like Roberts operator or Prewitts
operator? Show that the Sobel operator (3x3 mask) can be constructed by convolving a smoothing
mask (2x2) with a difference mask (2x2).

4. What is Laplacian of Gaussian and how do you compute it? What is the average value of LoG
operator and how do you obtain it? For simplicity you may consider 1D (one dimension/variable)
case. What is the average value of any image convolved with this operator and why?

5. Consider slope intercept representation of lines and give the main steps involved in performing
Hough transform for detecting lines in an image.

6. An image with a scene where it contains non-overlapping dark object on a light background is to
be segmented. The image however, has un-even illumination, which makes the image have more
illumination on the left side and less on the right side of the image such that left side looks brighter
than the right side of the image. How will you segment the image using thresholding?
 7. Why does thresholding not necessarily provide connected regions after segmentation? Explain
using a suitable example. In your example is it possible that a different threshold does provide
connected regions?

8. What is the meaning of optimal threshold while performing thresholding for segmentation? How
can you find an optimal threshold for bimodal histogram of an image? State your assumptions if
any.

9. What data structure would you use for performing split and merge of image regions for
segmentation? Justify use of this data structure using an example.

10. What is edge linking and why is it required? How is edge linking performed?

Unit 5: Image Compression

1. How do you find compression ratio? How is it related to the data redundancy?

2. A source of 6 symbols has probabilities: 0.40, 0.20, 0.15, 0.10, 0.10, 0.05. (i) Calculate the entropy
in bits per symbol. (ii) Construct the Huffman code for the symbols and calculate the average code
length in bits of the source per symbol. [Note: Please mention your assumptions and conventions
if any.]

3. What are the three basic data redundancies, which can be exploited for compression? Which one
of these can be used for loss-less compression?

4. An image consists of 8 independent gray levels {0,1,2,3,...,7}. The level 0 has a 0.5 probability,
and all the other remaining levels are equiprobable. What is the Entropy of this image (in
bits/pixel)? Construct a Huffman code for the gray levels assuming that you code each pixel value
separately. What is the average number of bits/pixel of the coded set of values?

5. Consider a source with L symbols {a1, a2, a3, aL} having associated probabilities as P(a1), P(a2),
P(a3)P(aL). Prove that the entropy H is bounded by What is the max H and in what conditions
one can attain the maximum entropy? For a two symbols source what is the maximum H and when
does one achieve it?

6. A monochrome digital image needs improvement in the contrast for which global histogram
equalization (HE) is to be used. The image has to be transmitted to a remote destination. You have
agreed to your seniors suggestion to use Huffman coding for compressing the image before
transmitting it. Now, you have to decide whether you do HE at the source or at the destination.
Justify your decision.

7. What is the code for a sequence of bab using arithmetic coding, where P(a) = 0.3 and P(b)=0.7.

8. It is observed that DCT in 1 d can be computed by obtaining the Fourier Transform of length 2N
of the even sequence f(n) which is extended from the input sequence g(n) of length N, i.e. Verify
the above observation. What are the advantages/disadvantages of using the above strategy?
 9. In what way the Lloyd-Max quantizer is optimal? Design the Lloyd-Max quantizer to find transition
and reconstruction levels for the input with uniform probability density function.

10. Why is DCT (Discrete Cosine Transform) a more popular transform for compression though KLT
(Karhunen-Loeve Transform) is optimal? Why is an image generally divided into sub-images
(blocks)? What is the typical block size considered in transform coding say using DCT and why?

11. Give the pipeline for the transform coding identifying each block.

12. For the baseline JPEG the following three types of images are given as input: A) An image of a
natural scene B) Image (A) with intensities of all the pixels divided by two, and C) a constant value
image. Assuming images are gray scale and of the same size, rank the images A, B, and C according
to the size of the JPEG compressed file (rank 1 for the largest size and so on). Justify your answer.

13. An image has some amount of salt and pepper noise. To remove the noise it is required that a
median filter is used. These images need up-sampling by a factor of two using bilinear interpolation.
The images also are to be compressed using JPEG. Following options are proposed for the sequence
of these three operations.

A. First up-sampling by two, then a median filter and then JPEG

B. First JPEG, then up-sampling by two and then median filter
C. First median filter then JPEG and then up-sampling by two
Rank these options (i) when the total number of bits in the compressed JPEG representation is
the least and (ii) when the artifacts and noise is the least. Justify your answer. [Note that JPEG
decoder is available if required in B and C]

14. What is the fundamental difference between Lossless and Lossy predictive coding? How do you
measure loss or error caused due to compression?

15. Suggest a loss-less method of compression for a binary image and give your reason as to why the
method proposed by you will do compression.