Correcting User Guided Image Segmentation

We tackle the problem of segmenting an image into planes given user input. Using a supervised learning approach, we develop an algorithm that adjusts the user input towards true plane segmentations, under the restriction that images have vertical boundary lines. We begin by collecting images for the training set consisting of hallways, buildings, stairs, rooms, etc. Users are then asked to draw dots on the image to indicate the locations of the intersection between neighboring planes. We then apply our supervised learning algorithm to predict the true location of the planes. Thus, we propose a method to infer the boundaries of planes from a single image given user input.