Spatial Filtering (Linear Filtering)

• Filters in Image Processing

• Correlation & Convolution
• Linear Filters

Dr.
Irfan
 Filters in Image Processing
□The name filter refers to modifying, or rejecting specified frequency components of an
image.

□For example, a filter that passes low frequencies is called a lowpass filter.

□Image processing filtersare mainly used to suppress either:

- The high frequencies in the image, i.e. smoothing the image

- The low frequencies i.e. enhancing/sharping or detecting edges in the image.

2
 What is Frequency in Images?
□ Frequency is the rate of change.
□ High- Frequency images:
□ The intensity of an image changes a lot, and the
brightness level change quickly from one pixel to the
next.

□ Low Frequency images:

□ Is relatively uniform in brightness or changes very Color changes gradually: Changes very rapidly
slowly A smooth, low- from one brightness to
frequency pattern. another: High-frequency
□
pattern.
Most of the images have both low and high-
frequency parts

3
 Fundamentals of Spatial Filtering

□ Spatial filters contains two components

□Mask / Convolution Matrix / Kernel: A neighborhood of pixel (typically small rectangle)
□Filter Function: A predefined operations that isperformed on the image pixels with the
help of neighborhood.

□ Filtering creates a new pixel value with coordinates equal to the

coordinates of the center of the neighborhood, and whose values is the
result of filtering operations.

4
 Fundamentals of Spatial Filtering

The above is repeated for every pixel in the original image to generate the filtered
image
5
 Spatial Filtering: Mathematical Form

□Spatial Filtering of an image of size𝑴 × 𝑵 with a filter of

size 𝒎 ×
𝒏 is given by expression

□ Where 𝒘 is the mask and 𝒇 isthe image function

6
 Spatial Filtering Fundamentals
□There are two closely related concepts that must be
understood clearly when performing linear sequential
filtering
□Correlations: process of moving a filter mask over the image and
computing the sumof product at each location

□Convolution: Same as correlation except the filterisfirstrotated by

180 degrees.

7
 Spatial Filtering Fundamentals

Correlatio
filter
n 244 510 738
244 255 246 1 2 3
1020 1200 1098
255 240 183 x 4 5 6 = = 8703

255 250 12 7 8 9 1785 2000 108

Convolutio Filter Rotated

n 244 255 246
180
9 8 7
2196 2040 1722

1530 1200 732

255 240 183 x 6 5 4 = = 10697
255 250 12 3 2 1 765 500 12

8
 Spatial Filtering: Mask / Kernel / Convolution Matrix
□ In image processing, a kernel, convolution matrix or mask is a small
matrix
□ Itisuseful for blurring, sharpening, embossing, edge detection and
more
□ Thisisaccomplished by means of convolution between a kernel
and an image.
□ Important Features of kernel
□ Size: 3x3,5x5,9x9,21x21etc.
□ Shape: rectangle, strip, circular or user defined
□ Coefficients /Values: set based on the operations
□ Anchor point: mostly in the middle
12
 Spatial Filtering Fundamentals: Convolution Matrix
□Convolution issimple mathematical
operation which is fundamental to
many common image processing
operators.

□Thiscan be used in image processing

to implement operators whose output
pixel values are simple linear
combinations of certain input pixel
values.

13
 Smoothing Spatial Filtering
□ One of the simplest spatial filtering operations, we can perform is a
smoothing operation.
□ Smoothing (averaging) spatial filters are used to reduce sharp transitions in
intensity (random noise typically consists of sharp transitions in intensity), ex: noise
reduction/blurring.

□ Types of smoothing filters:

□ Linear smoothing filters, ex:Gaussian filter

11
 Linear Noise Smoothing
□ The blurring (box, Gaussian) is a type of image-blurring filters that uses a
function for calculating the transformation to apply to each pixel in the
image.

□ Gaussian blur takes the weighted average around the pixel, while box-blur
just average all the pixels in the radius of the single pixel together.

12
 Linear Noise Smoothing – Gaussian Filter
□ Using OpenCV
import cv2
img = cv2.imread("pyimg.jpg")

blur_image = cv2.GaussianBlur(img, (7,7), 0)# (7,7): kernel

size- specifying height and width of Gaussian filter matrix,
must be odd numbers( 3*3, 5*5, 7*7), larger value create
more blurring, 0: sigma x(standard deviation in x direction)

cv2.imshow('Original Image', img) cv2.imshow('Blur Image',

blur_image)

cv2.waitKey(0)

13