Level Set Based Image Segmentation
This work is copyrighted.
Active contours without edges (ACWE)
In this work, image segmentation is carried out using the Active contours without edges (ACWE) framework. The ACWE method was proposed by Chan and Vese and is derived from the piecewise constant Mumford-Shah functional in a level set framework for image segmentation. The final Euler-Lagrange equations obtained after minimizing the functional are solved numerically using semi-implicit scheme. The review of the level set function, ACWE segmentation method, numerical solution and results are given in this pdf.
The coding is done in MATLAB.
The segmentation video for a grayscale image is shown below which has only 2 intensities - black and gray. Hence is called a "2 Phase" segmentation. This is the basic example which demonstrates the method.
The segmentation video for a grayscale image is shown below which has only 2 intensities - black and gray. Hence is called a "2 Phase" segmentation. This is the basic example which demonstrates the method.
Additional examples: The method can also be extended to multi-phase, multi-channel segmentation (Reference provided in the pdf). Examples are provided below. (Code, formulation not provided).
Example 1:
Multi-phase segmentation: Here 3 level set contours are used which can segment up to 8 regions (or phases) in the image. Below is a grayscale image consisting of 5 regions (where region intensities include different shades of grey). The video shows the evolution of the level set contours to segment the image.
Example 2:
Multi-phase, multi-channel segmentation: This is for images which have multiple channels like RGB images which have 3 channels namely: red, green and blue. In this example, a RGB image, which has 3 colors is segmented using 2 level set contours (which can segment up to 4 regions). The video shows the evolution of the level set contours for segmenting the image.