ECE 480 Lab 9

Ryan Eslinger and Mike Fuson

16 February 2011

Circles Erroded Circles Histogram of Circles

To count the Circles we used the ultimate point command in ImageJ which returns a pixel in the position of the original shape and gives the pixel a grey value equal to the radius of the shape. We then looked at the histogram of this output which told us that there were 3 circles with radius of 16, 7 with radius 17, 10 with radius 20 and 4 with radius 54.

Paragraph letter b hit-or-miss output

To count the occurrences of the letter b we used the following MATLAB code. At the end of the code it would output the variable count which equaled the number of occurrences. In this case it was 5.

% Michael Fuson

% Ryan Eslinger

% ECE480

% Lab9

clear;

clc;

img = im2single(rgb2gray(imread('black cat words.bmp')));

% hhist2d = video.Histogram;

% prob_dist = step(hhist2d,img).';

for i=1:283

for j=1:729

if img(i,j)>.5

img(i,j)=1;

else

img(i,j)=0;

end

y=1;

for i=17:77

x=1;

for j=143:185

b(y,x)=img(i,j);

x=x+1;

end

y=y+1;

end

figure

imshow(img)

figure

imshow(b)

output=bwhitmiss(img,b);

output=+output;

figure

imshow(output)

count = 0;

for j = 2:282

for i = 2:728

if output(j,i) == 1 && output(j+1,i+1) == 0 && output(j-1,i-1)== 0 && output(j-1,i+1) == 0 && output(j+1,i-1) == 0 && output(j+1,i) == 0 && output(j-1,i) == 0 && output(j,i+1) == 0 && output(j,i-1) == 0

count = count + 1;

else

end

count

We feel like we succeeded in this lab as we correctly counted the number of circles and letter b’s.

ECE 480 Lab 8

Ryan Eslinger and Mike Fuson

15 February 2011

We were not able to make the Huffman code work in Matlab but what we tried to do is the following. First we took the histories of the image then arranged the values according to their probability. We then added the least probable values and resorted the array. It was difficult to figure out how to track the resorting and making a tree that could then be used to produce the actual code. We would do this till the there were only two values left and then go down the tree to produce the code.

Lossless:

Start Lossless Code Lossless deCode

The decoded image was the exact same as the original. All detail was retained.

%Michael Fuson

%Ryan Eslinger

%ECE 480

%Lab8 Lossless

lena = im2single(rgb2gray(imread('lena.jpg')));

imshow(lena);

figure(1),title('Lena');

lenacode = lena;

for j = 1:512

pred = lena(j,1);

for i = 1:512

lenacode(j,i) = lenacode(j,i) - pred;

lenacode(j,1) = lena(j,1);

pred = lenacode(j,i);

end

figure(2)

imshow(lenacode);

title('LosslessLena');

lenadecode = lenacode;

for j = 1:512

predde = lenacode(j,1);

for i = 1:512

lenadecode(j,i) = lenacode(j,i) + predde;

predde = lenacode(j,i);

end

lenadecode(j,1) = lenacode(j,1);

end

figure(3)

imshow(lenadecode);

title('Lena Decode');

Lossy:

Start Lossy Code Delta = 0.1 Lossy deCode Delta = 0.1

Lossy Code Delta = 0.3 Lossy deCode Delta = 0.3

In both cases the decoded image was not the same as the original. Details were lost. The Step size of 0.1 was closer to the original when decoded.

%Michael Fuson

%Ryan Eslinger

%ECE 480

%Lab8 Lossy

lena = im2single(rgb2gray(imread('lena.jpg')));

imshow(lena);

figure(1),title('Lena');

lenacode = lena;

delt = .1;

for j = 1:512

pred = lena(j,1);

for i = 1:512

diff = lenacode(j,i) - pred;

lenacode(j,1) = lena(j,1);

if diff >= 0

pred = pred + delt;

lenacode(j,i) = delt;

else

pred = pred - delt;

lenacode(j,i) = -delt;

end

figure(2)

imshow(lenacode);

title('LosslessLena');

lenadecode = lenacode;

for j = 1:512

predde = lenacode(j,1);

for i = 1:512

lenadecode(j,i) = lenacode(j,i) + predde;

predde = lenadecode(j,i);

end

lenadecode(j,1) = lenacode(j,1);

end

figure(3)

imshow(lenadecode);

title('Lena Decode');

We feel that we successfully completed this lab.

ECE 480 Lab 7

Ryan Eslinger and Mike Fuson

26 January 2011

Compression:

For the Boats compression we used a threshold of 100 and for the Bridge we used the default threshold of 10. For the Boats image the file size went from 55KB to 46KB and for the Bridge image the file size went from 61 KB to 58KB. This shows that changing the threshold changes how much the image is compressed using the FSW module.

Boats Original Boats Compressed

Bridge Original Bridge Compressed

Filter:

For the filtering we used the FSW transform plugin. We did three iterations of the wavelet algorithm on each image and then removed the top left portion of the picture that represented an average of the picture. We then ran the process in reverse and obtained the images shown. The resulting images contain the edges of the picture which means that the process acted like a highpass filter. There was also a lot of ringing in the images which is indicative of a very sharp filter.

We also used different wavelets and repeated the process. For the different wavelets the intermediate pictures looked very different but the end results were very similar.

Boats Original Proc-Boats WT-Boats Boats 3iterations

Boats 3iterations filtered boats 3iterations reconstructed Boats with edges reconstructed

Bridge Original Proc-Bridge WT-Bridge

Bridge 3iteration Bridge 3iteration filtered

Bridge Orth Bridge B4 Bridge Dual

We feel that we successfully completed this lab.

ECE 480 Lab 6

Ryan Eslinger and Mike Fuson

19 January 2011

Benham’s Disc:

Below is a picture of our Benham Disk stationary and rotation. No color is visible on the picture in motion because the color effect is produced by the actual spinning of the disk. It is not complete understood why this pattern produces colors while moving. One theory is that the three different cones in the eye have different delays in their perception so somehow when the motion is at the right frequency color is perceived to appear

Stationary Spinning

Color Brightening:

RGB:

Original Stack Red Green Blue

Histogram Blue Equalized Histogram Blue

Equalized Blue Restack New Image

HSB:

Original Stack Hue Saturation Brightness

Histogram Brightness Equalized Histogram Brightness

Equalized Brightness Restack New Image

As learned in class, modifying just one color affects the other colors when it is restacked. The colors were changed. By modifying the brightness in the HSB components, none of the colors were changed and the image simply became brighter.

Pseudo Color Using Look Up Tables:

Original Tooth LUT New Tooth

We feel that we successfully completed this lab.

ECE 480 Lab 5

Ryan Eslinger and Mike Fuson

12 January 2011

Slow Blur Method and Results:

Stationary Image Blurred Image Deblurred Image

% Michael Fuson

% Slow clock deblurr

slowblurr = im2double(imread('clock moving slowly.bmp'));

imshow(slowblurr);

figure(1),title('clock moving slowly');

noise_var=2.5000e-005;

Len=55;

Theta=3;

PSF=fspecial('motion',Len,Theta);

estimated_nsr = noise_var / var(slowblurr(:));

Rslowblur = deconvwnr(slowblurr, PSF, estimated_nsr);

figure(2), imshow(Rslowblur)

title('Restoration of Blurred Image')

Rapid Blur Method and Results:

Stationary Image Blurred Image Deblurred Image

% Michael Fuson

% Ryan Eslinger

% ECE480

% Lab5

% Rapid clock deblurr

clc;

clear;

rapidblurr = im2double(imread('clock moving rapidly.bmp'));

imshow(rapidblurr);

figure(1),title('clock moving rapidly');

Theta=6;

noise_var=5.6e-5;

Len=115;

PSF=fspecial('motion',Len,Theta);

estimated_nsr = noise_var / var(rapidblurr(:));

Rrapidblur = deconvwnr(rapidblurr, PSF, estimated_nsr);

figure(115), imshow(Rrapidblur)

title('Restoration of Blurred Image')

We used Matlab to correct the degradation caused by the motion blur. Specifically we used the deconvwnr function which deconvolves an image using the wiener filter algorithm that we have learned about in class. To make the actual filter used to convolve with the image we used the fspecial command which has a specific parameter that allows it to approximate the degradation that occurs from camera motion. This is not totally ideal since the clock was moving and not the camera itself but it work sufficiently well. In the filter parameter we specify the amount the image was moving horizontally with the Len variable and the amount of rotational motion is specified by the theta variable.

We determined the Len and theta variables experimentally but running the code and comparing its output to the ideal image provided. For the slow moving clock we decided on a Len value of 55 and for the fast moving clock we decided on a Len value of 115. The theta values were small because the clock mostly had horizontal motion. Bellow you can see the resulting images and the specific code that we used to obtain them.

As mentioned in class it was very difficult to do this lab with great success. While we managed to deblur the slow moving clock fairly successfully, we found the rapid moving clock to be much more difficult. We do however feel that we did make an improvement to the rapid clock particularly around the T and the lower dial.

ECE 480 Lab 4

Ryan Eslinger and Mike Fuson

5 January 2011

Original Bridge

Large Pinhole Large Pinhole Convolution Large Pinhole Correlation

Small Pinhole Small Pinhole Convolution Small Pinhole Correlation

Two Pinholes Two Pinholes Convolution Two Pinholes Correlation

Dots Original Dots Correlation

Commentary On Part 2:

When we convolved the images with themselves we saw the pinholes that were originally used to make the images.

Commentary On Part 3:

The diameter of the circles was 43 pixels while the distance between the minima and the central maximum was 44 pixels.

We have completed the lab and believe we were successful in doing so.

ECE 480 Lab 3

Ryan Eslinger and Mike Fuson

15 December 2010

Original Image Original FFT

Filter to Remove Sinusoid Image After Sinusoid FFT After Sinusoid

Image After Median Filter FFT After Median Histogram After Median

Image After Enhance FFT After Enhance Histogram After Enhance

Image After Simple Median Only

First we took an FFT of the original image. We located the sinusoid and made a filter with two black circles at the location of the sinusoid. We blurred these black dots with a Gaussian filter to reduce ringing. We then did a custom FFT filter with our filter to remove the sinusoid. Next we did a median filter to remove the shot noise. Finally we enhanced the image using the equalize function.

We feel that the lab was successful as we removed the noise from the image.

ECE 480 Lab 2

Ryan Eslinger and Mike Fuson

8 December 2010

Histogram:

Histogram Before Histogram After

First we equalized the image.

Image Before Image After

We used a median filter to remove the shot noise.

The equalization spread the histogram and added contrast to the picture so that the image was easily visible. Once that was done there was lots of salt and pepper noise which we cleaned up using the median filter. This did cause some blurring.

Masks:

Original Image

Simple Average Mask Simple Average Image

1	1	1
1	1	1
1	1	1

Weighted Average Mask Weighted Average Image

1	2	1
2	4	2
1	2	1

Horizontal Sobel Mask Horizontal Sobel Image

-1	0	1
-2	0	2
-1	0	1

Vertical Sobel Mask Vertical Sobel Image

-1	-2	-1
0	0	0
1	2	1

Laplacian Mask Laplacian Image

1	1	1
1	-8	1
1	1	1

In order to add these masks we used ImageJ’s convolve option which is located under Process\Filters. We then simply typed in the filters that we wished to use. The simple average mask blurred the image. The weighted average mask blurred the

image and brightened the image.The Sobel masks detected edges in one direction. The Laplacian Mask detected all edges.

Fourier Image:

A) Here is the code we used:

// This macro creates a sine image

width = 512; height = 512;

xc = width/2; yc = height/2;

newImage("Sine wave", "32-bit black", width, height, 1);

for (y= 0; y<height; y++) {

for (x= 0; x<width; x++) {

setPixel(x, y, f1(x,y));

}

if (y%20==0) showProgress(y, height);