Illuminance Flow

Abstract

This thesis investigates the concept of illuminance flow, which is a novel topic to the area of computer vision and image analysis. Illuminance flow is a 2D flow field in the image produced by projecting the light vector into the local tangential plane of every visible point of a smooth object. The main application of illuminance flow is to monocular shape inference. Specifically, given a single image of an object, what is the shape of the visible surface of the object? In this sense, shape from illuminance flow is similar to shape from texture and shape from shading, but fundamentally independent of both. Illuminance flow is only observable from objects with rough 3D surface texture. Therefore, much of the work has been on 3D texture, and its special dependence on light direction, This thesis investigates the concept of illuminance flow, which is a novel topic to the area of computer vision and image analysis. Illuminance flow is a 2D flow field in the image produced by projecting the light vector into the local tangential plane of every visible point of a smooth object. The main application of illuminance flow is to monocular shape inference. Specifically, given a single image of an object, what is the shape of the visible surface of the object? In this sense, shape from illuminance flow is similar to shape from texture and shape from shading, but fundamentally independent of both. Illuminance flow is only observable from objects with rough 3D surface texture. Therefore, much of the work has been on 3D texture, and its special dependence on light direction, which is shown to be very different from “regular” 2D texture.