1. Multidisciplinary Engineering Senior Design Project P06441 See Through Fog Imaging Preliminary Design Review 05/19/06
Project Sponsor:
Dr. Rao
Team Members:
William Parsons, Philip Edwards
Team Mentor:
Kate Gleason College of Engineering
Rochester Institute of Technology

3. Objective
Characterize and optimize two different algorithms for removing fog
Each algorithm uses the same basic equation:
F(k)=((I(k)/C0 – 1)*exp(B*dist(k))) + 1
Equation affects image at the pixel level
Characterize parameters used to defog
B, C0, precision, threshold values
Compare speed, quality, and overall performance of both algorithms

4. Project Background
Currently small amount of data published on this process
Basic work to ensure current algorithms perform correctly
Only tested on artificially induced fog
No characterization of parameters

5. Needs Assessment Processing speed Quality of Image Overall Performance
For traveling applications speed is a major factor
Quality of Image
For surveillance applications quality should be very high
Overall Performance
Performance with different degrees of fog
Amount of noise each can handle
Accuracy of the constant predictions
B, C0

6. Key Requirements and Critical Parameters
High quality of defogged images
Low distortion
Processing time
Must produce images reasonable quickly B
Crucial to removing fog from images C0
Also affects the amount of fog removed
Early estimates suggest it is not as crucial as B

7. Design Concepts Color images Hardware Realization
Defog color images instead of black and white
Hardware Realization
Sample images from a camera, defog images, send to LCD screen
Analysis of Algorithms
Characterize parameters used in each algorithm
Optimize algorithms to increase overall performance

8. Chosen Design Analysis of the two algorithms
Not enough information known about other designs to make it feasible to design anything else
Determine good initial values for B and C0
Determine reasonable threshold values

9. Chosen Design
Determine reasonable precision needed to correctly approximate constants
Early tests suggest precision may need to be changed after each run through the algorithms
Determine differences in the two algorithms
Speed
Quality
Size of algorithm
Certain applications may only have a small amount of space available for algorithm

10. Algorithm Overviews
Each algorithm attempts to determine some constant values in the equation
B, C0 for first algorithm, B for the second
Each scales the pixels in the fogged image to be between zero and one
Guess at initial values and then change them to make particular pixel values converge
Smallest pixel value to zero and largest pixel value to one
Subtracts current constant values with previous values and compares with threshold values
If they’re smaller, constants have been found
Otherwise calculate new values

11. Overall System Design

12. Removing Induced Fog

13. Inducing fog in images
In order to successfully test the algorithms, fog must be artificially induced on images
Allows us to know correct B and C0 values
Do not need to determine distance vector, dist(k)
Can compare defogged image with original image
Visual inspection
Histograms
Frequency Analysis
etc

14. Inducing Fog

15. Anticipated Design Challenges
Determining correct amount of precision
Early tests lead us to believe precision will need to be changed each time through the algorithm
Determining distance vector in natural fog
Images where fog is not induced will have different distance vectors
Need to determine an algorithm to calculate the distance to each pixel
Determining threshold values for each constant
Testing with natural fog

16. Algorithms Precision
Algorithm comes close to approximating B and C0 values on first run
Successive runs show no improvement in either parameter
Precision value needs to be changed after each run through algorithm

17. Distance Vector Multiplied by B constant in both algorithms
Greatly affects transformation done on pixel values

18. Threshold Values
Determines when B and C0 have been successfully approximated
Need to determine when algorithms changes on the pixels is to small to notice
Also need to determine if this implies the constants are correctly approximated

19. Preferred Analysis Methods
Visual Inspection
Histogram
Frequency Analysis
Least Mean Square Error
Noise Sensitivity

20. Senior Design II
Parameter approximation should be close to actual values
Algorithms should run as quickly as possible
Defogging of natural fog should be implemented
Efficient way to determine distance vector in natural fog images should be found

21. Senior Design II Affects of different parameters should be determined
Comparisons of algorithms overall performance should be made
Processing Speed
Image Quality
Performance with differing amounts of fog

22. Questions?