<< Chapter < Page Chapter >> Page >
This test determines whether a fish is a salmon or a trout based on the detection of the number of fins it has using the 1-D DWT

So far the testing process could determine general color trends and size of the fish. Yet the fish size ratio test treated the fish as a block and fish contours have a good amount of detail in their fin patterns. Therefore, we decide that fins on a fish are important to examine.

We use the 2D DHWT on a gray intensity image and get back a low frequency image of the fish, for rangeto/2. We then look only at the rows of this signal that have enough value to be considered actually mass of the fishes body. For every row with mass, we transform the values into binary for the actual parts of the fish. We then take the derivative of that row vector which returns a positive 1 on a rising edge of a body and a negative 1 on the end of that body mass. The +1 points are more reliabe for a fish and give us the location of a fin. Yet, as we look at a fin that is slightly angled, we will get a diagonal matrix of 1s for the same fin. Therefore, we group +1 that are within a range and call that a fin.

Fin testing image

The binary fish image before the derivative is taken and the edges of the fins located.

Description of the discrete haar wavelet transform

As with all transforms, the DHWT takes the image signal and multiplies it by an orthonormal basis. Wavelets form a basis that are built by a mother wavelet and then different widths and shifts of that wavelet. The DHWT is built by averaging values next to each other and then taking the difference of the values. The Haar Wavelet Transform is ideal for low pass filtering and looking at different ranges of frequency. By taking different levels of the DHWT we can create different resolutions for an image, but each level of DHWT cuts the signal samples in half. We used the second level of DHWT to get similar blocks. DHWT also works as a rough edge detector but is not ideal.

Code for fin detection

% This function looks through a grayscale version of the input rgb image and % tries to detect fins by looking at each line of the picture and looking for% where the data starts and stops. function [fins]= findet(image); grayimage = rgb2gray(image);[wide, length] = size(grayimage);Mass = 0; n = 1;frontedge = zeros(1,(length/2 -1)); backedge = zeros(1,(length/2 -1));for i = 1:wide tempfishrow = dwt(grayimage(i, :),'haar');fishrow(i,:) = double(tempfishrow); if sum(fishrow(i,:))<= 0.9 Mass = Mass;else n = n + 1;Mass = 1 + Mass; fishrow(i,:) = fishrow(i,:) > 0;diffrow(i,:) = diff(fishrow(i,:)); for j = 1:(length/2 - 1)if diffrow(i,j) == 1 frontedge(1,j) = frontedge(1,j) + 1;elseif diffrow(i,j) == -1 backedge(1,j) = backedge(1,j) + 1;end endend endfrontedgefilter = frontedge >=10; frontedgeextreme = frontedge >=11;backedge = backedge >=3; for i = 1:(length/2 -1)if frontedgefilter(1,i) == 1 if sum(frontedgefilter(1, i:i+10)) > 1frontedgefilter(1, i:i+10) = [1 0 0 0 0 0 0 0 0 0 0];end endend fins = sum(frontedgefilter) - 1;
Got questions? Get instant answers now!

Get Jobilize Job Search Mobile App in your pocket Now!

Get it on Google Play Download on the App Store Now




Source:  OpenStax, Ece 301 projects fall 2003. OpenStax CNX. Jan 22, 2004 Download for free at http://cnx.org/content/col10223/1.5
Google Play and the Google Play logo are trademarks of Google Inc.

Notification Switch

Would you like to follow the 'Ece 301 projects fall 2003' conversation and receive update notifications?

Ask