Image Segmentation with Gaussian Mixture Model

Authors

Prasheel Renkuntla Shubham Sonawane Raj Prakash Shinde

Description

This project deals with the detection of underwater buoys of different colors using Gaussian mixture model. The Expectation Maximization algorithm is applied to learn the color distribution of the three buoys of colors Green, Orange and Yellow. The above procedure is first applied on 1D gaussian(single channel) then on 3D Gaussian distribution of the image.

Dependencies

• Ubuntu 16
• Python 3.7
• OpenCV 4.2
• Numpy
• matplotlib
• sys
• math

Run

To run the code for detecting all buoys using 3D gaussian

python3.7 model_all_colors_3D.py

Enter the choice of code (histogram, training or detection of buoys) when prompted

To run the code for 1D detection of yellow buoy

python3.7 yellow_detection.py

To run the code for 1D detection of green buoy

python3.7 green_detection.py

To run the code for 1D detection of orange buoy

python3.7 orange_detection.py

Demo

First, we create a dataset from each of the frames of the video and distinguish each colored buoy. A sample orange buoy is given below-

Cropped orange buoy

Based on the average histogram for each colored buoy we decide the number of cluster that would be needed to identify the color distribution. The gaussian distribution for orange colored buoy is given below-

Gaussian distribution of Orange Buoy

Now, after color distribution is known, we use these parameters to detect the buoy individually first, then all three combined on the input video. The top left corner image is the mask obtained from weights learned, the bottom left corner is the edges that we determine from the mask. The next output is the top right image of the dilated version of edges. From this we find the mean and make a circle around it to detect the buoy as below-

Pipeline for buoy detection at 140th frame

The final image below shows the progressive detection-

Buoy detection for yellow, orange and green buoys

The Output folder has video results from 3D GMM only.

Reference

• https://cmsc426.github.io/colorseg/#colorclassification
• https://towardsdatascience.com/gaussian-mixture-models-explained-6986aaf5a95
• https://towardsdatascience.com/gaussian-mixture-modelling-gmm-833c88587c7f