GMMImgSegmentation

A tiny project using GMM to segment imgs. This is done by:

1. First convert the image from RGB color space into RG color space
2. Then assume the pixels to be points from K different multiple dimentional Gaussians
3. Apply EM methods to obtain the fitted data points

Example picture

K=3 Case

Picture converted into rg space

Greyscale pictures reflect how likely(the possibility) that a pixel may lay in one Gaussian component.

• White: More likely
• Black: Less likely

Color distribution and Gaussians fitted

K=4 Case

Greyscale pictures

Color distribution and Gaussians fitted

K=5 Case(notice the over-fit)

Greyscale pictures

Color distribution and Gaussians fitted

+Notice the tiny green located in blue area