SEGMENTING IMAGES THROUGH REPRESENTATION AS A CONTENT-RICH NETWORK
Yanira Corvera, Laura Smith.
California State University Fullerton, Fullerton, CA.
Image segmentation is the process consisting of partitioning digital images into multiple sections with the goal of detecting objects within the image. In this work, we represent an image as a graph, where each pixel is a node in the graph and an edge is present between 2 nodes if they are close spatially within the image. In particular, we treat the image as a content-rich graph, where each node is associated with a vector of different features of the image, such as RGB and HSV (hue saturation value) color representations and textures. We consider graph cut and clustering methods to partition the nodes, resulting in a segmented image. We use the information theoretic approach, the content map equation, which minimizes the description length of the network by compressing the graph into distinct modules of nodes giving the partitioning of the image. We compare this method to existing graph-cut methods, such as normalized cut, on a variety of images. Implementation of the content map equation on a simple black and white image provides promising results which encourage further investigation on more complex images.