P06641 Continuation Results
Table of Contents
This experiment is designed to determine how the algorithm reacts to errors within the distance matrix. This is done by setting up a distance matrix within the fogging algorithm and then adding/subtracting from each value within the matrix at the beginning of the algorithm. This was done for each of the images and values ranging between -100 and 100 was added to the matrix values.
B and C0 Iteration Testing
This experiment is designed test how quickly the algorithm converges on the correct B and C0 values. Ten starting values will be chosen and the B and C0 values will be recorded during each iteration through the algorithm. This is accomplished by setting a breakpoint at the beginning of the algorithm and recording the values of B and C0 each time the algorithm reaches the breakpoint.
Combined Mean Square Error Results
Data from previous tests will be combined to create a comparison of the calculated results which will aid in determining patterns in B, C0, and the iteration testing previously completed.
Histogram Stretching vs. Algorithm One
Will be used to determine which performs better using the mean square error of the resulting images. Each will be used to defog multiple images with different initial B and C0 values and the resulting Mean Square Error of both will be compared. The algorithm should perform better then histogram stretching. Several different degrees of fog will be used ranging from small amount to large amount, roughly twenty different images.
stretch(r,c)=(orig(r,c) - val) * val2
This experiment is designed to test the affect of noise on the convergence of the algorithm. This is accomplished by adding different amounts of noise density to the original image and checking the resulting MSE, B, C0, and the resulting image itself via visual inspection. The amount of noise added varied from .001 to 1 using the salt and pepper noise addition formula in Matlab outlined at:
Natural Fog Testing
Images will be taken with no fog and the distances to objects in the image will be measured. These measured distances will be used to construct the distance matrix used in the defogging algorithm. Images with natural fog from these locations will be taken and ran through the algorithm using the measured distance matrix. The resulting image will be compared to image with no fog to determine the accuracy of the algorithm
If images with natural fog cannot be taken, fog will be induced using the measured distance matrix to attempt to closely imitate natural fog.
The mean square error and visual inspection of the defogged image and the fogged image will be compared in order to determine the effectiveness of the algorithm.