[Python] Proper Curve fitting for Intensity Plots

I am doing image analysis by analysing the histogram of intensity values. I get similar plots (2 sharp peaks and one flat peak in specific order).

I tried to model it using mixture of Gaussian but the results are not good. I have attached the images for reference. After BIC Scores analysis, I found that the knee is when I set the number of gaussians to 4.

Out of 5 cases tried, following 2 gave good results:





However, following 3 were incorrect in terms of their results:







Are there alternate/better methods to approximate the curves?

I tried changing the gaussian parameters, or suppressing the intensities sharpness by applying average filter over the image and then taking the plots. Still did not help.

I tried approaches in the following: Curve fitting for n detected peaks and different intensity

I am looking for 3 different curves one for the two peaks that appear towards the end, one for the plateau like part in the starting and maybe 1 curve with high variance and low weight for the other noise stuff.

Any solution that uses traditional CV Methods OR the methods in machine learning are most welcome.

Edit: Source of histograms I am analysing the histograms that were generated after segmenting the intestines from a segmentation model on 3D CT Scan Images. For almost 100 images, I get the similar curves on manual inspection. The first plateau is fat, the first peak ahead is the distribution of contents and the further next peak is the distribution of intestinal wall. X axis represents a linear function of HU (hounds field unit) values.

Continue reading...