MATLAB: How to compare two images from each other: Round 2

image processingimage tracking

Hello again!
Slightly stuck on this issue. I need help on finding a way for MatLab to follow an object from one picture to the next, automatically. I've written a small script from help online that allows me to select the object and determine its coordinates, but is there a way that MatLab can see the object without me manually tracing it?
Thanks!,
-Frank
Images:
Code:
    I = imread('rice.png');
    %%imshow(I)
    background = imopen(I,strel('disk',15));
    figure, surf(double(background(1:8:end,1:8:end))),zlim([0 255]);
    set(gca,'ydir','reverse');
    %%Darkens image
    I2 = I - background;
    %figure, imshow(I2)
    %%Increases contrast
    I3 = imadjust(I2);
    %figure, imshow(I3);
    %%Turns image into binary colors (Black and White)
    level = graythresh(I3);
    bw = im2bw(I3,level);
    bw = bwareaopen(bw, 50);
    %%figure, imshow(bw)
    cc = bwconncomp(bw, 4);
    cc.NumObjects
    %%Select grain connected to background
    grain = false(size(bw));
    grain(cc.PixelIdxList{50}) = true;
    figure, imshow(grain);
    labeled = labelmatrix(cc);
    whos labeled
    %%Color Shift
    RGB_label = label2rgb(labeled, @autumn, 'c', 'shuffle');
    figure, imshow(RGB_label)
    graindata = regionprops(cc, 'basic');
    graindata(1).Area
    %%Grain area
    grain_areas = [graindata.Area];
    [max_area, idx] = max(grain_areas);
    grain = false(size(bw));
    grain(cc.PixelIdxList{idx}) = true;
    figure, imshow(grain)
    %%Select a portion to allocate coordinates of object
    [x, y, ~, xi, yi] = roipoly(grain);
    BW = roipoly(100, 200, xi, yi);
    %%Displays xi, yi coordinates 
    figure, imshow(BW)
    grain2 = imread('Rice3.gif');
    [x2,y2,Bw,xi2, yi2] = roipoly(grain2);

Best Answer

For this rice example, what do you know about that grain that makes it different that the rest?
  • Its size (+- a few pixels due to digitization)
  • The area of its boundingbox
  • Eccentricity
These are just three options of:
doc regionprops
that could be used.
  • Basically you would isolate the object you want in the first image.
  • Run the same binarization operation on the second image
  • do a connected components analysis in both, keeping the connected component of interest in the first one. (doc bwconncomp)
  • Get the measurements of all objects in the second image.
  • Whichever one is the most similar is your match.
Note: these are just a few of the things you could do. Looking at your above binary image, binarized, they seem feasible. If you want to compare in non-binary images then we might have to figure out something a little more complicated.
By the way, well written question - this is what allows you get a good answer
Related Question