VRL an eye

Variational Image Segmentation using Color and Texture


Baris Sumengen, C. Kenney, B.S.Manjunath


This segmentation method uses image diffusion based on an edge vector field. If the image is indicated as a texture image, Gabor texture features are calculated at 3 scales and 4 orientations. After image features are calculated/retrieved (texture, color, grayscale), they are smoothed. An edge vector field is calculated from these features using directional differences. These vectors point towards the closest edges such that image diffusion is towards the edges (similar to shock filtering). In the next step, an edge function is calculated from this vector field by solving a Poisson equation. Last step is the image diffusion.


The following demonstrates the steps taken to create the color+texture segmentation of a lizard image:

Initial Image (384x256)

Flowfield (Click on the image for full view [3456x2304])

Edge Function

Start image diffusion - Diffused image

Segmentation Result