Automatically crop an object from a solid background as close as possible to the object edges.

I am trying to crop an object/product from a relatively solid background but have been struggling to get a perfect implementation or at least something that works most of the time.

My previous questions were asked on Stack Overflow and I've been able to get decent results and cropping a rectangle works almost perfectly:

matlab-crop-a-polygon-from-an-image

subtract-background-from-image

matlab-object-detection-using-edge-detection-and-bounding-boxes

matlab-auto-crop

My current approach is to find a mask of the image and then change every unwanted pixel to white, but this hasn't given me consistent results:

original = imread('1.jpg');
level = graythresh(original);

img = rgb2gray(original);
mask = im2bw(img,level+0.1);

se = strel('line',10,90);
mask = imdilate(~mask,se);

mask = imfill(mask,'holes');

bw = activecontour(img,mask);

rows = numel(original(:,1,1));
columns = numel(original(1,:,1));

for i = 1:rows
    for j = 1:columns
        if (  bw(i,j,1) == 0 )
            original(i,j,:) = 255;
         end
    end
end

imshow(original);

A preview image:

Answer 1

[Image Analyst]

From the looks of your code, it looks like you should come to Answers before Stack Overflow. For example, you last double loop can be done vectorized. Anyway, if you want as close as possible, why are you dilating your image and doing an activecontour? Both of these will give you a boundary slightly off your "true" boundary. I'd recommend converting to hsv colorspace and threshold for high saturation. The objects will be selected and the white background will not be selected (because white and shadows have low S value). See the color segmentation demos in my File Exchange, especially this one: http://www.mathworks.com/matlabcentral/fileexchange/28512-simple-color-detection-by-hue

[Jacques Fourie]

Hi Image Analyst, I get the mask pretty well, but how do I get rid of any writing inside the product label and how do I use the mask to isolate the product, I tried the following code and the valueMask and saturationMask looks good on their own...

rgbImage = imread('5.jpg');

% Convert RGB image to HSV
  hsvImage = rgb2hsv(rgbImage);
  % Extract out the H, S, and V images individually
  hImage = hsvImage(:,:,1);
  sImage = hsvImage(:,:,2);
  vImage = hsvImage(:,:,3);

    hueThresholdLow = 0;
    hueThresholdHigh = graythresh(hImage);
    saturationThresholdLow = 0.09;
    saturationThresholdHigh = 1.0;
    valueThresholdLow = graythresh(vImage);
    valueThresholdHigh = 1.0;

  hueMask = (hImage >= hueThresholdLow) & (hImage <= hueThresholdHigh);
  saturationMask = (sImage >= saturationThresholdLow) & (sImage <= ... saturationThresholdHigh);
  valueMask = (vImage >= valueThresholdLow) & (vImage <= ... valueThresholdHigh);

      mask = (saturationMask & valueMask);