how to create rounded edges of an irregular object? I want to smoothen the corners of this object.

Answer 1

[Image Analyst]

You can blur the image with imfilter() or conv2():

windowSize = 21; % Bigger for more blurring.
kernel = ones(windowSize) / windowSize^2;
output = imfilter(rgbImage, kernel);

[Nabanita Roy]

Thank you for the answer. I could apply this with little modification. However, the bwarea is more after smoothning than the original binary image..How can I can smoothen the corners by eroding pixels. I have one more question. How to get maximum inscribed circle or ellipse for this image? I am using Matlab R2019a.

[Image Analyst]

It depends on how you did your segmentation. Perhaps you want to blur after it's been segmented:

windowSize = 21; % Bigger for more blurring.
kernel = ones(windowSize) / windowSize^2;
threshold = 0.5; % Adjust to change size of new mask
mask = conv2(double(mask), kernel, 'same') > threshold;
area1 = nnz(mask) % Area in terms of number of pixels.
area2 = bwarea(mask) % Area using a wegihting scheme.

Note that because its shape is different, it will have a different area than the unsmoothed image.

To get the max inscribed circle, you can get the distance transform using bwdist(~mask). Then find the max value of it. That will be the radius. The location of the max is your circle center. Something like (untested):

edtImage = bwdist(~mask);
imshow(edtImage, []);
radius = max(edtImage(:))
[centerRowY, centerColumnX] = find(edtImage == radius);
centers = [centerColumnX, centerRowY]
viscircles(centers, radius);

[Nabanita Roy]

Thank you so much for your support. Is there any way to smoothen the boundary micro projections of the object by erosion dilation?

[Image Analyst]

Yes, you can use imclose(). It does a dilation, which smoothes the boundary, then does an erosion of the same amount to return it to it's original size, just with a slightly different (smoother) shape. For large kernel sizes, this process will tend to have more artifacts than blurring and thresholding the binary image like I suggested. But try it and see. Good for learning if you try different things.

[Nabanita Roy]

Thanks for your suggestion.

[Nabanita Roy]

I have multiple same images. Is it possible to crop all the binary images in such a way that the black object will take only 40 percent of the entire image and the rest 60 percent will remain white. Thats how I want to standardize the images.

[Image Analyst]

Yes. First get the area, then divide by 0.4 to get the number of pixels in the cropped image. Then use indexing to crop it out.

mask = bwareafilt(mask, 1); % Take largest blob only.
props = regionprops(mask, 'area', 'Centroid');
imageArea = area / 0.4;
imageWidth = sqrt(imageArea);
xCentroid = props.Centroid(1);
yCentroid = props.Centroid(2);
xLeftCol = round(xCentroid - imageWidth/2);
xRightCol = round(xCentroid + imageWidth/2);
yTopRow = round(yCentroid - imageWidth/2);
yBottomRow = round(yCentroid + imageWidth/2);
output = mask(yTopRow : yBottomRow, xLeftCol : xRightCol);