2D-to-3D Conversion System Based on Linear Perspective and Scene Classification

• Main Authors: Ken-Min Lin, 林肯民
• Other Authors: Ching-Huang Wei
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/93146006955262239364

碩士 === 國立高雄第一科技大學 === 電腦與通訊工程研究所 === 102 === The general 2D-to-3D conversion system first estimates depth map from 2D image, and utilizes the estimated depth map combined with the original 2D image to generate a stereo image through the depth image based rendering (DIBR) method. Therefore, depth map plays an important role in 2D-to-3D conversion system, but a 2D image contains little depth information. The thesis focuses on estimating the structure of linear perspective in image using Hough Transform, a line detection algorithm to find vanishing line and vanishing point and to analyze the linear perspective of the whole scene. Besides, the objects in 2D image are segmented using 4-connected component labeling technique, and depth value are given to the different objects to generate a final depth map. In order to make sure that the 3D image keeps the visual quality and comfortable viewing, the estimated depth map is smoothed using joint bilateral filter. Finally, original 2D image is combined with depth map to generate 3D image using depth image based rendering method.