NATIONAL INSTITUTE OF TECHNOLOGY, ROURKELA – 769008
END-SEMESTER EXAMINATION, 2017
SESSION: 2017-2018 (Autumn)
Ph.D. + M. Tech. 1 Semester (ESC) + Dual Degree 9th Semester+ [Link].
st
Subjec Code: EE653 Subject Name: Digital Image Processing Dept. Code: EE
No. of pages : 2 Full Marks : 50 Duration: 3 hrs
Qs. Answer all questions. marks
Illustrate your answers with neat sketches as and when necessary.
1. a. An image having non-uniform illumination distribution has to be segmented for some 4
application. Construct an efficient algorithm based on grayscale thresholding method which
can effectively segment the image. Justify your suggestion.
b.
Hough Transform is a global edge linking method applied after detection of edge pixels. 3
Elaborate the steps for identifying the edge lines of an image using the above.
c. Show that the two dimensional Gaussian operator is separable while the Laplacian of 3
Gaussian operator is not separable.
2. a.
Assume that P1 and P2 are the a priori probabilities of object and background pixels of an
4
image, where p1(z) = (z-1) for 0 < z < 3 corresponds to the intensity of the objects and p2(z)
= (z-2)/2 for 2< z < 4 corresponds to the intensity of the background. Assuming that P1 =
3/4 and P2 =1/4, derive and find the optimal threshold intensity value between object and
background pixels.
b. Canny’s edge detection approach need to do two important steps, i.e. nonmaximal 4
suppression followed by hysteresis thresholding. Elaborate these steps with the objectives.
c. Suppose B is a binary image and J, K are two different structuring elements as shown 2
below.
J = K=
Please explain the purpose of the operation (B ϴ J )∩ ( Bc ϴ K ), where ϴ is the erosion
operator.
3. a. In two-band sub band coding, design the analysis and synthesis filter banks such that the 5
system behaves as a perfect reconstruction filter. Find the relationship between the filter
coefficients. Hence design a 2D four band filter banks for sub band image coding.
b. 5
Compute the one-dimensional Discrete Wavelet Transform of function f(n) = { 1,-5, 4, 2}
for 0 <= n <= 3 with base layer j0= 0. Use Haar scaling and Haar wavelet functions for
computation of the coefficients. Reconstruct the original function from its transform. Sketch
the samples and used functions.
4.a. In video compression, motion estimation and compensation is an important step. Draw the 4
block diagram of the motion compensated video encoder. Elaborate the steps to perform
the predictive video coding.
b. How many computations are required to find the optimal motion vector of an 8X8 3
macroblock using the MAD optimality criterion, single pixel precision, and a maximum
allowable displacement of 8 pixels? What would it become for ¼ pixel precision?
c.
From the multi resolution analysis of a signal, derive the scaling function co-efficient and
3
wavelet function co-efficient. Hence generate the refinement equation and its orthogonal
complement equation.
5.a. Describe the steps of the JPEG 2000 image compression algorithm. Indicate what kind of 4
redundancy (i.e., coding, interpixel, psychovisual) is addressed by each step. Why is DWT
is used instead of some other transform? Which step causes information loss?
3
b. Show that geodesic erosion and dilation are duals with respect to set complementation.
Assume that the structuring element is symmetric about its origin.
c. Apply mask B to the following image A to perform the dilation operation. Show the result 3
after one iteration.
A = B=
