Evaluation of image segmentation without using a model

Question

Hend Abouche 2022-5-12

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1718125-evaluation-of-image-segmentation-without-using-a-model

评论： Image Analyst 2022-5-13

Hello guys, I've been trying to see who is better at image segmentation is it using only the green channel mask the whole image mask

here is my code for segmentation using green channel

I=imread(image);
g=I(:, :, 2);
%binarizing images
gg=imbinarize(rescale(g));
SE=strel('disk',5);
%green
gg=imopen(gg,SE);
gg=imdilate(gg,SE);
gg = bwareafilt(gg,1);
maskedRgbImageg= imoverlay(I,gg,'k');

my segmentation code using the whole image

I=imread(image);
gray=rgb2gray(I);
%binarizing images
graym=imbinarize(rescale(gray));
SE=strel('disk',5);
graym=imopen(graym,SE);
graym=imdilate(graym,SE);
graymmaskedRgbImageg= imoverlay(I,graym,'k');

and now I don't know how to evalute each one segmentation using the metrics, is there is any methode that can show me who segments better. thank you in advance

2 个评论
显示无隐藏无

DGM 2022-5-12

编辑：DGM 2022-5-12

在 MATLAB Online 中打开

jagblobs.png

That's entirely dependent on the image, the content, and the intent. It's not clear what those are, and it's not clear what "the metrics" are either.

Just grabbing an offhand color image,

I get this:

I = imread('jagblobs.png');

g = I(:, :, 2);

%binarizing images

gg = imbinarize(rescale(g));

SE = strel('disk',5);

gg = imopen(gg,SE);

gg = imdilate(gg,SE);

gg = bwareafilt(gg,1);

maskedRgbImageg = imoverlay(I,gg,'k');

imshow(maskedRgbImageg)

... and this

I = imread('jagblobs.png');

gray = rgb2gray(I);

%binarizing images

graym = imbinarize(rescale(gray));

SE = strel('disk',5);

graym = imopen(graym,SE);

graym = imdilate(graym,SE);

graymmaskedRgbImageg = imoverlay(I,graym,'k');

imshow(graymmaskedRgbImageg)

Neither of which really help to make the intent clear (the approximate removal of the largest object, maybe?).

Hend Abouche 2022-5-13

yes I am using the bwareafilt function to keep only the largest object

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Image Analyst 2022-5-13

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1718125-evaluation-of-image-segmentation-without-using-a-model#answer_963340

I would use whatever image gave you the most contrast in what you want to find. Usually a single color channel will give you the most contrast if you're looking for something that is a single color. If you want to find multiple colors simultaneously then useing rgb2gray would probably be best. You might even be able to use PCA to get more contrast. PCA Demos attached.