The Applications of Segmentation to Indoor Scenes

Chun-Jung Tai, Purdue University

Abstract

In this thesis, we present an application of detecting Stegatone and two applications of segmentation to indoor scenes. In the first topic, we studied a number of factors that affect how effective a periodic peak can be detected. The rest of the thesis discusses two applications of segmentation to indoor scenes. The first application is related to a click based segmentation on indoor screens. The next application discusses the lattice detection and segmentation on near-regular tiles. An extended topic is shown in the appendix which discusses two different lattice fitting error functions and its estimators. Stegatone, an encryption technology based on printing the halftone dots. The detection of quasi-periodic patterns, such as those found in clustered-dot halftones, can be efficiently achieved by searching for strong peaks in the frequency domain. In this thesis, we quantify four factors that contribute to the attenuation of those characteristic peaks related to mobile hand-held image capture. These include MTF, halftone cluster size, blur, and contrast. We derive the expected theoretical attenuation for each of these factors, and then compare these with experimentally measured results from mobile captured images of test prints. We present a click-based interactive segmentation for indoor scenes, which allows the user to select an object or region within the scene in a few clicks. The goal for the click-based approach is to provide the user with a simple method to reduce the amount of input required for segmentation. We first present an effective global segmentation strategy, which provides a rough separation of different textures. The user, then, places a few clicks to segment the target. A novel trimap assignment strategy is proposed to utilize the click information. To study the performance of our method, psychophysical experiments were conducted to compare our click-based approach with other existing methods. In interior design, tiles are one of the most commonly used materials. We propose an interactive lattice detection and segmentation approach for near-regular tiles in order to virtually re-render the appearance. The objective of this chapter is to automatically detect the layout of near-regular tiles and to segment the grout based on a bounding box indicated by users. The algorithm applies an initial segmentation based on Gaussian mixture model after which, the dominance peaks are computed as a proposal of the lattice vectors. By ranking the combination of vectors among the top dominance peaks, the correct periodicity matrix is determined. Then, a local adjustment of lattice points is proposed. We apply graph cut with an estimated trimap to segment the tiles from grout. Our result shows that the lattice can be accurately found with an average error less than 5%. More over, we address the skew of classes by evaluating the result with ROC and Precision-Recall curves. Finally, we investigate the impact of tile image degradation on the accuracy of the lattice estimation algorithm and the accuracy of the segmentation algorithm.

Degree

Ph.D.

Advisors

Allebach, Purdue University.

Subject Area

Electrical engineering

COinS