How to Crop Multiple Objects in an Image [MATLAB] - matlab

Question

I'm freshman to MATLAB & Developing "Rice Quality Identification" Application using MATLAB & NEURAL NETWORK .For my Guidance I'm preferring this Research Paper
This Application Comprises with 5 Phases
Image Acquisition
Image Pre-processing
Image Segmentation and Identifying Region of Interest
Feature Extraction
Training and Testing
I'm now in 3rd Phase , Already developed initial steps for this application
Step 1: Browse Image from Computer and Show it
% Get the orginal image & show , Figure 1
[fileName, pathName] = uigetfile('*.jpg;*.tif;*.png;*.gif','Select the Picture file');
I = fullfile(pathName, fileName);
I = imread(I);
imshow(I)
Step 2: Background subtraction
% selected rice image Background subtraction , Figure 2
% Use Morphological Opening to Estimate the Background
background = imopen(I,strel('disk',7));
I2 = I - background;
figure, imshow(I2);
Step 3:
% get the Black and white Image , Figure 3
% output image BW replaces all pixels in the input image with luminance greater than 0.17 level
BW = im2bw(I2,0.17);
figure, imshow(BW)
Step 4:
% Remove small objects fewer than 30 pixels from binary image
pure = bwareaopen(BW,30);
figure, imshow(pure)
Step 5: Labeling
% Label Black and white & Image bounding box around each object
L=bwlabel(pure,8);
bb=regionprops(L,'BoundingBox');
I'm Sticking at Step 6 since 2 days. Step 6 is crop multiple objects from original image using Labeled Binary Image
which is exactly output should get like below image ,
if I can get this I can easily calculate Morphological Features and Color features for each object in that original image , to use for phase 4 .
Morphological Features
1.Area for each Object
2.scale of X, Y axis for each object in above picture
3.using X, Y axis I can Calculate Aspect Ratio
Color features
1. Red Mean
2. Green Mean
3. Blue Mean
Can you please explain the way to crop multiple objects from original image using Labeled Binary Image which is Step 6.

Answer 1

If I am interpreting Step #6 right, I believe what it's saying is that they want you to segment out the final objects after Step #5 using the binary map that you have produced. Given your comments, you also want to extract the bounding boxes delineated in Step #5 as well. If that's the case, then all you have to do is use the RegionProps structure defined in bb that will help us do this for you. As a bit of review for you, the BoundingBox field of a RegionProps structure for each object extracted from the image returns an array of 4 numbers like so:
[x y w h]
x denotes the column / horizontal co-ordinate, y denotes the row / vertical co-ordinate, and w,h denote the width and height of the bounding box.
All you need to do is create a binary map, and cycle through each bounding box to delineate where we need to cut out of the image. When you're done, use this binary map to extract out your pixels. In other words:
%//Initialize map to zero
bMap = false(size(pure));
%//Go through each bounding box
for i = 1 : numel(bb)
%//Get the i'th bounding box
bbox = bb(i).BoundingBox;
%//Set this entire rectangle to true
%//Make sure we cast off any decimal
%//co-ordinates as the pixel locations
%//are integer
bbox = floor(bbox);
bMap(bbox(2):bbox(4), bbox(1):bbox(3)) = true;
end
%//Now extract our regions
out = zeros(size(I));
out = cast(out, class(I)); %//Ensures compatible types
%//Extract cropped out regions for each channel
for i = 1 : size(out,3)
chanOut = out(:,:,i);
chanIm = I(:,:,i);
chanOut(bMap) = chanIm(bMap);
out(:,:,i) = chanOut;
end
This creates an output image stored in out, and copies only those pixels that are true from each channel over, based on each bounding box given in Step #5.
I believe this is what Step #6 is talking about. Let me know if I have interpreted this properly.