JSS Science and Technology University, Mysuru

Master of Computer Applications

Topic: Role of Statistics and Probability theory in CVPR
Presented by: Gajanand Patil, Jagannath V. Bhandari, Sushanth V. S
Table 1. Various statistical measures and their formulas
1. Role of Statistics 1.1.2 Median 2.3 Probability distribution function
Measure Formula
•Median is measure of intensity level of pixel which is separating the Arithmetic Mean •The Cumulative Distribution Function F(x) is F(x) = P(X ≤ x)
•Statistics is the study of the collection, organization, analysis, and
interpretation of data. It deals with all aspects of this, including the
high intensity value pixels from lower intensity value pixels. It is a type •The Probability Mass Function for a discrete variable is P(x) = P(X = x)
planning of data collection in terms of the design of surveys and
of order-statistic filter. The most popular and useful of the rank filters •The Probability Density Function for a continuous variable is the
Geometric Mean
is the median filter. function P(x) such that
experiments.
•It works by selecting the middle pixel value from the ordered set of P(X ≤ x) = ∫ x−∞ P(u)du
•The various statistical measures are mean, mode, median, variance,
values within the ‘m×n’ neighborhood ‘W’ around the reference pixel. Fig 1A. Original Image Fig 1B. Image with Fig 1C. Image after AM
Harmonic Mean We write P(X = x) for the probability distribution (density or mass) that
standard deviations, covariance, skewness and kurtosis. Gaussian noise filtering of Fib 1B
This filter simply sorts all values within a window, finds the median random variable X takes value x.
•The application of these measures in the various fields of digital image
value, and replaces the original pixel value with the median value.
processing like, image enhancement, image restoration, image Contra-Harmonic Mean
•It is commonly used for salt and pepper noise.
denoising, and edge detection etc 2.3.1 Bernoulli Distribution
1.1.3 Mode Median •X is a random variable that either takes the value 0 or the value 1.
1.1 Application: Image Denoising •Let P(X = 1|p) = p and so P(X = 0|p) = 1 − p.
•In statistics , the mode is the value that occurs most frequently. In •Then X has a Bernoulli distribution.
image processing filter will each pixel value by its most common Mode
Fig 1D Image after GM Fig 1E. Image with Salt
1.1.1 Mean neighbor. filtering of Fig 1B & Pepper noise
Fig 1F. Image after HM
filtering of Fig 1B 2.3.2 Binomial Distribution
•This is a particularly useful filter for classification procedures where Standard Deviation
• Mean is most basic of all statistical measure. Means are often used each pixel corresponds to an object which must be placed into a class;
•The binomial distribution is obtained from the total number of 1’s in n
in geometry and analysis; a wide range of means have been in remote sensing, for example, each class could be some type of
independent Bernoulli trials.
developed for these purposes. terrain, crop type, water, etc.
Skewness •X is a random variable that takes one of the values 0, 1, 2, . . . , n.
•In contest of image processing filtering using mean is classified as
•Let P(X = r|p) = ( n/r ) pr (1 − p)(n−r) .
spatial filtering and used for noise reduction. 1.1.4 Standard Deviation •Then X is binomially distributed.

A. Arithmetic Mean •It is a most widely used measure of variability or diversity used in Kurtosis
2.4 Application: Image brightness
statistics. Fig 1G. Image after
Fig 1H. Image after
•The arithmetic mean filter also known as averaging filter, operates on •In terms of image processing it shows how much variation or Contra-HM filtering of Fig 1I. Image after
median filtering of
an sliding ‘m×n’ window by calculating the average of all pixel values dispersion exists from the average (mean, or expected value).
Fig 1B
Fig 1E
mode filtering of Fig1B
2.4 1 Probability distribution function of the
within the window and replacing the center pixel value in the •As it has capability in measuring the variability, it can be used in edge
destination image with the result.
brightness
•The arithmetic mean filter causes a certain amount of blurring
sharpening, as intensity level get changes at the edge of image by large 2.2 Application: Image Enhancement •The probability distribution function P(a) , is the probability that a
value.
(proportional to the window size) to the image, thereby reducing the •Standard deviation filters can be useful for radar images. The •Given a two-dimensional image { x}jk shown in Fig 2A, we define the brightness chosen from the region is less than or equal to a given
effects of noise and local variations. It can be used to reduce noise of interpretation of radar images is often difficult: you cannot rely on random variable X as the pixel values of the image with the sample brightness value a.
different types, but works best for Gaussian, uniform, or Erlang noise. spectral values because of back scatter. space of X = {x0=0, x1=1, x2=2,..., x255=255}. The image has a probability •As a increases from -∞ to +∞ , P(a)increases from 0 to 1. P(a) is
density function f X (x) monotonic, non-decreasing in a and thus dP|da>=0.
B. Geometric Mean • We would like to transform the image to have the probability density
1.1.5 Kurtosis function fY(y)=(1/64)e-y/64.
2.4 1 Probability density function of the
Fig 1J. Image after Fig 1K. Image after Fig 1L. Image after

•The geometric mean filter is a variation of the arithmetic mean filter •In statistics, kurtosis is a measure of the shape of the probability
standard deviation skewness filtering of kurtosis filtering of Fig •As we have seen, the transform is performed by replacing each {x} jk brightness
filtering of Fig 1A Fig 1A 1A with a new value {y}jk , which will depend both on fx (x) and fy (y) . The
and is primarily used on images with Gaussian noise. This filter is distribution of a real-valued random variable. It is closely related to the •The probability that a brightness in a region falls between a and a+∆a,
sample space of Y = {y0=0, y1=1, y2=2,..., y255=255}. given the probability distribution function P(a) can be expressed
known to retain image detail better than the arithmetic mean filter. fourth moment of a distribution. A high kurtosis distribution has •Assuming a continuous random variable Y, the desired cumulative
longer, fatter tails, and often (but not always) a sharper peak. A low 2. Role of Probability density function is given by FY ( y ) = ∫0y fY (ω)dω = 1− e− y 64 .
as P(a)∆a ,where P(a) is the probability density function.
kurtosis distribution has shorter, thinner tails, and often (but not
C. Harmonic Mean always) a more rounded peak.
•It follows that
FY −1 (z)= −64log(1− 0.989z),
•The harmonic mean filter is yet another variation of the arithmetic
•In digital image processing kurtosis values are interpreted in 2.1 Random variables where z is treated as a uniformly distributed random 3. References
combination with noise and resolution measurement. High kurtosis number with 0 ≤ z ≤1.
mean filter and is useful for images with Gaussian or salt noise. Black values should go hand in hand with low noise and low resolution. •A random variable, usually written X, is a variable whose possible •Thus the image transformation we require is given by •E. Kreyszig Advanced Engineering mathematics, J. Willey & Sons Inc.
pixels (pepper noise) are not filtered.
values are numerical outcomes of a random phenomenon. {y}jk = −64log(1− 0.989FX ({x}jk )). 2011.
1.1.6 Skewness • There are two types of random variables, discrete and continuous. •The image that results from applying this operation to the sample •R. Gonzalez, R.E. Woods, Digital Image Processing, Upper Saddle River,
D. Contra-harmonic Mean •A discrete random variable is one which may take on only a image in Fig 2A is shown in Fig 2B NJ, Prentice Hall, 2002.
•In statistics, skewness is a measure of the asymmetry of the countable number of distinct values such as 0,1,2,3,4,........ Discrete •Charles W. Therrien and Murali Tummala, Probability for Electrical and
•The contra-harmonic mean filter is another variation of the arithmetic probability distribution of a real-valued random variable. The skewness random variables are usually counts. If a random variable can take only Computer Engineers, CRC Press, 2004.
mean filter and is primarily used for filtering salt or pepper noise (but value can be positive or negative, or even undefined. a finite number of distinct values, then it must be discrete •Roy D. Yates and David J. Goodman Probability and Stochastic
not both). Images with salt noise can be filtered using negative values •In terms of digital image processing, Darker and glossier surfaces tend •A continuous random variable is one which takes an infinite number Processes: A Friendly
of R, whereas those with pepper noise can be filtered using positive to be more positively skewed than lighter and matte surfaces. Hence of possible values. Continuous random variables are usually •Introduction for Electrical & Computer Engineers, John Wiley
values of R. we can use skewness in making judgements about image surfaces. measurements. Examples include height, weight Fig 2A. Fig 2B. Publishers, 2005. Chapter 3,section 3.7.