Imaging systems (e.g. cameras) often require accurate, automatic, and fast estimates the lens' focus error. Maximally accurate focus error estimates are obtained by first calculating the pattern of contrast at each level of detail (i.e. spatial frequency) in individual natural image patches (e.g. 64x64 pixels). Next, using Bayesian statistics, we determine how probable each focus error is by comparing the contrast pattern in each patch to contrast patterns that are characteristic of different focus errors. The characteristic patterns depend on the properties of the imaging system's optics, sensors, and noise. They characteristic patterns also depend on the properties of imaged natural scenes. In a typical smart phone camera, chromatic aberrations can be used to correctly estimate the sign of the error 98% of the time. Estimates are accurate, precise, and can be computed in ~1 millisecond.
Patent / Licensing Questions
Optimal focus error estimation performance for the optics of a Samsung Galaxy phone.
• Burge J & Geisler WS (2014). Optimal focus error estimation performance in individual images of a popular smart phone.
Technical Document, University of Texas at Austin. [ pdf ]
• Burge J, Geisler WS (Dec. 2013) An image-based method for the optimal estimation of defocus in individual natural
images. Patent Publication No. US-2013-0329122-A1 [ pdf ]
• Burge J, Geisler WS (Feb. 2011) An image-based method for the optimal estimation of defocus in individual natural
images. Provisional Patent 22084-P069.
• Burge J & Geisler WS (2012). Optimal defocus estimates from individual images for autofocusing a digital camera.
Proceedings of the SPIE, 8299, 82990: E1-E12, January: Burlingame, CA [ pdf ] (Best Paper Award!)
• Burge J, Geisler WS (2011). Optimal defocus estimation in individual natural images. Proceedings of the National
Academy of Sciences, 108 (40): 16849-16854 [ pdf ]