Refocusing images captured from a stereoscopic camera

ABSTRACT

Traditional photography projects a 3D scene to a 2D image without recording the depth of each local region, which prevents users from changing the focus plane of a photograph once it has been taken. To tackle this problem, Ng et al. [2005] presented light-field cameras that record all focus planes of a scene and synthesized the refocused image using ray tracing. Nevertheless, the captured photographs are of low resolution because the image sensor is divided into sub-cells. Levin et al. [2007] embedded a coded aperture on the camera lens and recover depth information from blur patterns in a single image. However, the coded aperture blocks around 50% of light. Their system requires longer exposition time when taking pictures. Liang et al. [2008] also embedded a coded aperture on the camera lens to capture the scene but with multiple exposures. It produces high quality depth maps yet is not suitable to hand-held devices. Recently, Microsoft Kinect directly estimates the depth information using infrared light, which works only in a indoor environment.

References

1. Hirschmuller, H. 2008. Stereo processing by semiglobal matching and mutual information. IEEE Trans. Pattern Anal. and Mach. Intell. 30, 2, 328--341.
2. Levin, A., Fergus, R., Durand, F., and Freeman, W. T. 2007. Image and depth from a conventional camera with a coded aperture. ACM Trans. Graph. 26, 3, 328--341.
3. Liang, C.-K., Lin, T.-H., Wong, B.-Y., Liu, C., and Chen, H. 2008. Programmable aperture photography: Multiplexed light field acquisition. ACM Transactions on Graphics 27, 3, 55:1--55:10.
4. Ng, R., Levoy, M., Bredif, M., Duval, G., Horowitz, M., and Hanrahan, P. 2005. Light field photography with a handheld plenoptic camera. Stanford University Computer Science Tech Report CSTR 2005-02.

Index Terms

• Computing methodologies

  • Artificial intelligence

    • Computer vision

      • Image and video acquisition

        • 3D imaging