Semi-auto 2D to 3D image conversion apps

Abstract

3D effects are becoming more and more popular, especially after the release of Movie Avatar in 2010. Recently, 3D display hardware is available such as 3D TVs and 3D monitors and become more and more. Due to the release of some glasses-free 3D smartphones, it arouses interest of some people to convert some 2D image or photos into 3D version. However, there is lack of applications of 2D to 3D image conversion on mobile applications. In this project, a 2D to 3D image conversion Android application is developed. It contains 2 modes, one is auto mode and the other one is semi-auto mode. In auto mode, the horizontal translation is used and it is suitable for the image that contains single object with simple background, for example, portrait image. In semi-auto mode, it uses 1) image segmentation for depth estimation and 2) Depth Image Based Rendering for 3D image synthesis. This mode can convert most 2D image into 3D version with a better results and user needs to draw different colours on the image to represent how far the object(s) is. For the Image segmentation, the watershed segmentation is used to separate the object and assigned the depth value. For the Depth Image Based Rendering, it consists of 2 parts, one is 3D image wrapping and the other is hole filling. It is used to combine the source image with the corresponding depth map in order to generate the 3D synthesis. Using these methods, the semi-auto mode produces a good quality of 3D image.