Area/Mask Processing Methods

(Trucco, Chapt 3)

- A pixel’s value is computed from its old value and the values of pixels in its vicinity.

- More costly operations than simple point processes, but more powerful.

• What is a Mask?

- A mask is a small matrix whose values are called weights.

- Each mask has an origin, which is usually one of its positions.

- The origins of symmetric masks are usually their center pixel position.

- For nonsymmetric masks, any pixel location may be chosen as the origin (depending
on the intended use).

1
1 1 1 1 2 1
1
1 1 1 2 4 2
1
1 1 1 1 2 1
1
• Applying Masks to Images (filtering)

- The application of a mask to an input image produces an output image of the same
size as the input.

Convolution

(1) For each pixel in the input image, the mask is conceptually placed on top of
the image with its origin lying on that pixel.

(2) The values of each input image pixel under the mask are multiplied by the
values of the corresponding mask weights.

(3) The results are summed together to yield a single output value that is placed
in the output image at the location of the pixel being processed on the input.
 -2-

Area or Mask Processing Methods

input image enhanced image

z1 z2 z3
z4 z5 z6 z5’ g(x,y) = T[f(x,y)]
z7 z8 z9 T
T operates on a
neighborhood of pixels

w1 w2 w3 z5’ = R = w1z1 + w2z2 + ... + z9w9

w4 w5 w6
w7 w8 w9

- Mathematical definition of -discrete- convolution:

n/2 n/2
g(i, j) = Σ Σ h(k, l) f (i − k, j − l)
k=−n/2 l=−n/2

(for a mask with odd dimensions)

 -3-

Cross Correlation

- Correlation translates the mask directly to the image without flipping it.

- It is often used in applications where it is necessary to measure the similarity

between images or parts of images.

- If the mask is symmetric (i.e., the flipped mask is the same as the original one)
then the results of convolution and correlation are the same.

n/2 n/2
g(i, j) = Σ Σ h(k, l) f (i + k, j + l)
k=−n/2 l=−n/2
 -4-

Non-linear filtering

- Linear filters have the property that the output is a linear combination of the
inputs.

- Filters which do not satify this property are called non-linear.

- Erosion and Dilation are examples of non-linear filters.

• Normalization of mask weights

- The sum of weights in the convolution mask affect the overall intensity of the result-
ing image.

- Many convolution masks have coefficients that sum to 1 (the convolved image will
have the same average intensity as the original one).

- Some masks have negative weights and sum to 0.

- Pixels with negative values may be generated using masks with negative weights.

- Negative values are mapped to the positive range through appropriate normalization.

• Practical problems

- How to treat the image borders?

- Time increases exponentially with mask size.

 -5-

Smoothing (or Low-pass) filters

- Useful for noise reduction and image blurring.

- It removes the finer details of an image.

• Averaging or Mean filter

- The elements of the mask must be positive.

- The size of the mask determines the degree of smoothing.

 -6-

• Gaussian (linear filter) (from "Machine Vision" by Jain et al., Chapt 4 and Trucco Chapt 3)

- The weights are samples from a Gaussian function.

(example using 1D Gaussian)

−(x 2 + y 2 )
h(x, y) = exp[ ]
2 2

- The Gaussian’s mask weights fall off to (almost) zero at the mask’s edges.

- Gaussian smoothing can be implemented efficiently thanks to the fact that the
kernel is separable:

n/2 n/2
g(i, j) = Σ Σ
k=−n/2 l=−n/2
h(k, l) f (i − k, j − l) =

n/2 n/2 −(k 2 + l 2 )

Σ Σ exp[ 2 2 ] f (i − k, j − l) =
k=−n/2 l=−n/2

n/2 −k 2 n/2 −l 2
Σ exp[ 2 2 ] l=−n/2
k=−n/2
Σ exp[ 2 2 ] f (i − k, j − l)
 -7-

Algorithm
To convolve an image I with a nxn 2D Gaussian mask G with
=

1. Build a 1-D Gaussian mask g, of width n, with

g =

2. Convolve each column of I with g, yielding a new image I c

3. Convolve each row of I c with g

- The value of
determines the degree of smoothing.

- As increases, the size of the mask must also increase if we are to sample the

Gaussian satisfactorily.

height = width = 5 (subtends 98.76% of the area)

 -8-

• Median filter (non-linear)

- Effective for removing "salt and pepper" noise (random occurences of black and
white pixels).

- Replace each pixel value by the median of the gray-levels in the neighborhood
of the pixels
Area or Mask Processing Methods

input image enhanced image

10 20 20
15 99 20 20 g(x,y) = T[f(x,y)]
20 25 20 T
T operates on a
neighborhood of pixels

10 20 20 15 99 20 20 15 20

sort
10 15 20 20 20 20 20 20 99

median
 -9-

Sharpening (or High-pass)

- It is used to emphasize the fine details of an image (has the opposite effect of
smoothing).

- Points of high contrast can be detected by computing intensity differences in local

image regions.

- The weights of the mask are both positive and negative.

- When the mask is over an area of constant or slowly varying gray level, the result of
convolution will be close to zero.

- When gray level is varying rapidly within the neighborhood, the result of convolu-
tion will be a large number.

- Typically, such points form the border between different objects or scene parts (i.e.,
sharpening is a precursor step to edge detection).

input image mask

−1 −1 −1
10 10 10 10 10 10 80 80 80 1/9 x −1 8 −1
10 10 10 10 10 10 80 80 80 −1 −1 −1
10 10 10 10 10 10 80 80 80
10 10 10 10 10 10 80 80 80
10 10 10 10 10 10 80 80 80

1/9 (−10 − 80 − 80 −10 + 640 −80 −10 −80 −80) = 210/9 > 0
(there is variation in the gray−levels)

1/9 (−10 − 10 − 10 −10 + 80 −10 −10 −10 −10) = 0

(there is no variation in the gray−levels)
 -10-

• Sharpening using derivatives

- Computing the derivative of an image has as a result the sharpening of the

image.

- The most common way to differentiate an image is by using the gradient.

 ∂f 
grad( f ) =  ∂x 
 ∂f 
 ∂y 

- The gradient can be approximated by finite differences which can be imple-

mented efficiently as masks.

- Examples of masks based on gradient approximations with finite differences:

 1 −1   −1 1 
   
 −1 1   1 −1 

Prewitt masks
 −1 −1 −1   −1 0 1
 0 0 0   −1 0 1
   
 1 1 1   −1 0 1

My Mx