You are on page 1of 3

Digital Image Processing

Sample Questions [Ch - 10, 11 & 12]

1. What is the significance of Laplacian of Gaussian operator in edge detection?


2. Explain Canny edge detector.
3. Explain the method of getting a circle in a grayscale image using Hough
transformation.
4. What are the strengths and limitations of graph theoretic segmentation
techniques? Find the edge corresponding to the minimum cost path in the sub
image shown. Assume the edge starts in the first row and ends in the last row.
1 0 5
2 4 4
0 7 6
5. What are the limitations of basic global and adaptive thresholding?
6. How to achieve optimal threshold value for the segmentation of a non-uniformly
illuminated image having two object classes? Derive it.
7. Considering that the object classes have Gaussian densities with different means
and variances find the optimal threshold values in terms of a priori probabilities
P1 and P2.
8. Suppose that an image has the gray level PDF shown. Here P1(z) corresponds to
objects and P2(z) corresponds to the background.
 1 x ≤ 20

P1(z) = (35 − x) / 5 20 < x < 35
 0 x ≥ 35


 0 x ≤ 20 or x ≥ 60

P2(z) = ( x − 20 ) / 15 20 < x < 35
(60 − x ) / 15 45 < x < 60

1 35 ≤ x ≤ 45
Assume that P1=0.4 and P2 =0.6. Find the optimal threshold between object and
background pixels.
9. Explain object segmentation via region merging and splitting.
10. Discuss Watershed segmentation algorithm for segmentation of a grayscale
image. What is the limitation of Watershed segmentation algorithm? How to
overcome it?
11. Give a step by step implementation of the dam construction procedure for a given
image segment A = {0, 2, 3, 7, 1, 1, 4, 0, 2, 8, 2, 5, 0}.
12. Explain different types of polygonal approximation techniques used to
approximate digital boundaries.
13. How convex deficiency does simplify the boundary description process? Why
should we use polygon approximation prior to the process of getting convex
deficiency?
14. Explain chain codes as a boundary representation technique? Provide a descriptor
that uses chain code representation technique and is translation, rotation and scale
invariant.
15. What do you mean by MAT of a region? Provide a thinning algorithm to get the
skeleton of a binary image. Justify that this algorithm yields the skeleton of a
binary image.
16. What is the significance of using Fourier descriptors? Summarize the Fourier
descriptors which are translation, rotation, scaling and starting point invariant.
17. How to exploit statistical moments for boundary description?
18. How can we make use of topological properties for region description? Provide
the topological descriptors for the shapes of the characters 0, 1, 8, 9 and X in
conjunction with convex hull.
19. Provide the histogram based statistical approaches to quantify textures in an
image. What is the limitation of these types of measures? To overcome this
limitation, how can we make use of co-occurrence matrix?
20. Obtain the gray level co-occurrence matrix of a 5X5 image composed of a
checkerboard of alternating 1’s and 0’s if the position operator P defined as “ one
pixel to the right” and if defined as “two pixels to the right”. Assume that top left
pixel has value 0.
21. Explain structural and spectral approaches for texture description.
22. Given a 4X4 image whose gray levels ordered lexicographically are as follows: 2
3 0 1 1 3 1 2 0 2 0 3 1 1 2 3. Calculate the normalized spatial moments up to the
second order.
23. How to use Principal Component Analysis for segmented region description?
24. Given a matrix A =
1 2 1
2 3 4
3 4 2
Compute |A|, A-1, Trace of A, Euclidean Norm of A, Eigen values and Eigen
Vectors of A.
25. Explain relational descriptors. Give a spatial relationship and corresponding tree
representation for a checkerboard pattern of black and white squares. Assume that
the top left element is black and that the root of the tree corresponds to the
element. The tree can have no more than two branches emanating from each node.
26. Explain minimum distance classifier in short. Find the decision boundary between
the classes w1 having the pattern vectors {[1, 1.3]T, [2, 0.7]T, [1.4, 3]T} and class
w2 having the pattern vectors {[5, 3.8]T, [6, 2.7]T, [6.4, 6]T}.
27. How to use correlation for pattern recognition?
28. How to implement Bayes classifier for pattern recognition? Suppose in a two
class problem the samples are from Gaussian distribution. The mean vectors are
as follows: m1= [2, 3]T and m2 = [5, 7]T. Assuming a priori probability of class
w1 and w2 are 0.4 and 0.6 respectively. Find the class boundary.
29. Why should we use a neural network for any classification problem? Explain the
Perceptron learning algorithm for a two class problem.
30. Apply Perceptron algorithm to the training set w1:{[0 0 0]T, [1 0 0]T, [1 0 1]T, [1 1
0]T} and w2: {[0 0 1]T, [0 1 1]T, [0 1 0]T, [1 1 1]T}. Let c=1 and w(1)= [-1 -2 -2
0]T. Find the decision boundary.
31. Give the architecture of a multilayer feed-forward neural network for pattern
classification? Explain how to train the network, using delta learning rule, through
back propagation.
32. Explain two basic structural techniques for pattern recognition.
33. Provide a syntactic method for structural pattern recognition.
34. How to use a finite automaton as a string recognizer?
35. Explain tree based pattern recognition.

You might also like