How to apply region growing for candidates?

I have been working on the diabetic retinopathy and currently implementing a paper named " Detection of microaneurysms using multi-scale correlation coefficients " by Bob Zhang and others.The link for the paper is :https://www.sciencedirect.com/science/article/abs/pii/S003132031000004X

I have to apply region growing for the candidates(the white spots present in the above image) so that the output should look like the below image where the candidates shape and size are grown by region growing.

In the paper, It is mentioned that In the region growing, the background image Ibg can be obtained by applying a median filter of size 25x25 to green channel image Igreen. A threshold 't' is calculated using the equation

t = Idarkest - 0.5(Idarkest - ibg)

where Idarkest denotes the lowest intensity for each candidate region in the Igreen, ibg is its background intensity at the same location.

Region growing starts from the point of Idarkest in each candidate region and continues until no more connected pixels are higher than threshold. Considering the size of MA is less than 120 pixels if the area of every resultant connected component is larger than 120 pixels, it will be discarded.

Can somebody help me with this?

I have attached the green channel image Igreen as 'green_sample.jpg'.

[Rik]

Have a read here and here. It will greatly improve your chances of getting an answer.

I don't see how this is related to region growing. If you need a region growing function, you could use this function, but I don't think that is actually what you want to do.

I'm not planning on reading the paper to understand your question, nor do I think many others are likely to. So I would suggest making your question separate from the paper. Of course you can provide a link for context and attribution, but it shouldn't be necessary to read the paper to understand your question.

[KALYAN ACHARJYA]

Have you applied the region growing on the first image (Binary Image)?? Why?

[Krishna Chaitanya]

@KALYAN ACHARJYA As per the paper mentioned above, The White dots in the first image represent the candidate microaneurysms that need to be detected for the diagnosis of Diabetic retinopathy.However, the size and shape of these candidates may not be same as that of the true microaneursysms and hence region growing is applied to the whitedots in the first image to get output as in second image.

Answer 1

[Image Analyst]

See my attached regiongrowing program. Adapt as needed.

[Krishna Chaitanya]

I'm new to matlab.Can you please help me with applying it to my requirement?Thank you.

[Image Analyst]

OK, but I'm traveling this week and don't have much time, certainly not enough time to completely program up some paper for you or do the whole thing for you. What have you done so far? Do you have a short question about some part of your adapted code?

[Krishna Chaitanya]

I have two images 1)a binary image with the white dots(first image in the question) and

2)a retinal fundus image of same shape (from which the binary image is obtained.)

i'm trying to figure out a way to find a pixel in each candidate( white region in binary image) such that its corresponding intesity value in the fundus image should be the least compared to that of other pixels in the same white region.

In this way,I need the indices of the least intensity pixel and the intesity value of every white region so that i can compute a threshold to apply region growing.

I would be grateful if you provide some insights on how to do this task.

Thak you.

[Rik]

There are probably more efficient methods, but you could do it like this:

1. use bwlabel to label all white regions
2. use second output to loop through all labeled areas
3. in each iteration, use L==k as a binary mask to index into your image and use mean, min, median, max, whatever to find the pixel intensity you need.

[Krishna Chaitanya]

Thank you Rik!!