Question

Answer the following questions Q1. The image digitisation process consists of, in general, three components to convert the continuous signal into digital form, including sampling (or discretisation in time/space), quantization (discretisation of amplitude) and coding (generation of binary code for each quantised level) a. Name the major distortions associated with the above image digitisation process [2 marks] Name the areas of principal applications (or the basic classes of problems) which are covered by image processing. b. Discuss the main issues and objectives of each of the above named areas [5 marks] Given a digital image shown in Figure Q1.1, discuss the effects of processing the image by a two-dimensional (2-D) low-pass filter. c. If the given digital image is represented by a 2-D matrix, x, and a 2-D low-pass filtered image is denoted by x/p = Tip{x} where Tip represents the 2-D low-pass filtering operation, comment on the following resultant images in terms of their frequency contents and visual appearance: 1.) the difference image, xdif= x - Xip» and 2.) a processed image, xproc X+ Xdiff [3 marks] 0 IANS WAY TO PEDESTRIANS Figure Q1.1 A high definition (HD) digital image (a right view frame from d. A monochrome (i.e., grey-scale) image, x, whose height and width are M and N respectively, is corrupted by one of the noise signals as specified in (Q1.d.i) and (Q1.d.ii) Advise what filtering technique can be used to remove the noise and to restore the image for each of the aforementioned two cases; and explain why. The noise signals are modelled as: i. An random impulse noise, n, with its impulse magnitude distributed randomly in the range of [0, 255] such that the corrupted image, xorrupt ?, is defined as (n[m,n] with probability Pm (Q1-1) Хсоmтрr Lm,n )- m,n] with probability 1-P where 0sms M-1, 0S»SN -1 and p. denotes the noise ratio (i.e., the percentage of the corrupted pixels in a given image) [I mark A two-di ii. sional (2-D) noise consisting of two 2-D sinusoidal signals, N1 and N2, at different spatial frequencies, i.e., A N,[M/2+u,N/2+v] = N}[M/2-u, N/2-v]={ ifu=u, and v = v otherwise (Q1-2 and Aifuu, and v= v2 ,N/2- ]={0 N2IM/2+u, N/2+v]=N2[M/2-u,N otherwise (Q1-3) where N and n2 represent the sinusoidal signals in the frequency domain, A1 and A2 are finite positive values, 0

Answer 1

Applications of Image Processing
Visual information is the most important type of information perceived, processed and
interpreted by the human brain. One third of the cortical area of the human brain is dedicated
to visual information processing.
Digital image processing, as a computer-based technology, carries out automatic processing,
manipulation and interpretation of such visual information, and it plays an increasingly
important role in many aspects of our daily life, as well as in a wide variety of disciplines and
fields in science and technology, with applications such as television, photography, robotics,
remote sensing, medical diagnosis and industrial inspection.
• Computerized photography (e.g., Photoshop)
• Space image processing (e.g., Hubble space telescope images, interplanetary probe
images)
• Medical/Biological image processing (e.g., interpretation of X-ray images, blood/cellular
microscope images)
• Automatic character recognition (zip code, license plate recognition)
• Finger print/face/iris recognition
• Remote sensing: aerial and satellite image interpretations
• Reconnaissance
• Industrial applications (e.g., product inspection/sorting)

objectives of image processing

Understand differences between computer vision and image processing.
Know the basic components of an image processing system.
Understand the basics of the human visual system as they relate to image processing; including spatial frequency resolution and brightness adaption.
Understand how images are represented; including optical images, analog images, and digital images. Understand image types such as binary images, gray-scale images, color and multi-spectral images.
Know the key concepts in image file formats.
Understand the model for an image analysis process.
Understand why preprocessing is performed and know about image geometry, convolution masks, image algebra and basic spatial filters.
Understand image quantization in both the spatial and brightness domains.
Understand how discrete transforms work; including concepts of basis images, orthogonality, orthonormality, separability and reversibility.
Know about the 2-D Fourier, discrete cosine, Walsh-Hadamard and wavelet transforms; including implied symmetry, phase, circular convolution, vector inner and outer products and filtering.
Know why log remapping is necessary for viewing spectral image data.
Understand lowpass, highpass, bandpass, notch filters; including ideal and non-ideal filters such as the Butterworth.

Non-Trivial world - Consider Segmentation problem !
Accuracy - You will never get satisfied with the results obtained ! (I have never heard 100% results in Image processing)
Hard Coded / No generic solutions - Consider a filtering problem or a contrast enhancement problem which needs to be solved generic but varies according to many factors (source, kernel, methods, etc) resulting in hard coding.
Diversified - More number of ways and many possible solutions for a problem. For instance you need boundary of a object -> start with edge detection (you have around 5 edge detection algorithms) and end up somewhere near skeletonization or distance transform.
All-Time Research - Start to work an application -> it is obvious in IP you are not going to attain 100% accuracy -> continue to improve the accuracy !
Etc ...Etc ... ( end of thinking capability :-P )

Some of well deserved researches are (personally I have used / known to me)
1. Viola Jones Face detection ( This undergoes all the above mentioned issues - But still outperforms other methods available)
2. HOG -Human / Pedestrian Detection
3. Convolution neural networks - Think Facebook uses this method for its Face recognition which surpassed human capabilities
4. Part Based Models - Improved method of (2)Hog features for Human / Pedestrian detection