Você está na página 1de 1

Barrett - CS450 Winter 2011

Lecture 8-9 Spatial filtering


Objectives: To understand the use of convolution and filter masks to implement a variety of
spatial filters.
See accompanying powerpoint slides.
Also, as an excellent supplemental resource on Image Arithmetic see:
http://homepages.inf.ed.ac.uk/rbf/HIPR2/filtops.htm
1. A running average is just a simple filter:
Signal:

3-point average:

2. A weighted average is also just a simple filter:


Signal:
3-point weighted average:
(weights: .25, .5, .25)

4.5

5.75

2.5

1.5

2.25

3. Both Running and Weighted Averages can be implemented as a Convolution


So whats a convolution? Just what we did multiply-add-sum (Ratchet analogy)
Strictly speaking, convolution means reverse weights before multiplying
4. Adobe Photoshop Demo with input of different weights
Try a lot of different weights under Filters > Custom (with and without scaling and bias)
Smoothing with Uniform weighting (box filter)
Smoothing with Gaussian weighting with radius of different sizes
- How can this do a 100x100 Gaussian Smooth so quickly? (Separability into 2 1-D smooths)
Edge Detection (Horizontal, Vertical) then combine (with Opacity) to illustrate
Gradient Magnitude
Laplacian
Sharpening
5. Median filter
Zoom image up create white pixel, black pixel over small ROI right over the top of an edge.
Then use Median Filter to clean it up without destroying the edge.
6. Unsharp Masking
Reverse-engineer the Unsharp Masking Kernel.
7. The Laplacian Operator
Derive the Laplacian Operator as the difference of the differences.

Você também pode gostar