High Dynamic Range Imaging Cecilia Aguerrebere Advisors: Julie Delon, Yann Gousseau and Pablo Musé Télécom ParisTech, France Universidad de la República, Uruguay
High Dynamic Range Imaging (HDR) Capture a scene containing a large range of intensity levels... Limited dynamic range of the camera loss of details in bright and/or dark areas.
High Dynamic Range Imaging (HDR)... using a standard digital camera. Limited dynamic range of the camera loss of details in bright and/or dark areas.
High Dynamic Range Imaging (HDR)... using a standard digital camera. Limited dynamic range of the camera loss of details in bright and/or dark areas.
High Dynamic Range Imaging (HDR)... using a standard digital camera. Limited dynamic range of the camera loss of details in bright and/or dark areas.
High Dynamic Range Imaging (HDR)... using a standard digital camera. Limited dynamic range of the camera loss of details in bright and/or dark areas.
High Dynamic Range Imaging (HDR)... using a standard digital camera. Limited dynamic range of the camera loss of details in bright and/or dark areas.
HDR Image generation HDR generation Irradiance Map (number of photons reaching each pixel per unit time)
HDR images in current technology Smartphones Standard digital cameras (e.g. Pentax K-7 DSLR, Canon PowerShot G12, Canon PowerShot S95) HDR generation HDR visualization (tone mapping)
HDR images in current technology Smartphones Standard digital cameras (e.g. Pentax K-7 DSLR, Canon PowerShot G12, Canon PowerShot S95) HDR cameras HDR DISPLAY HDR No need of tone mapping HDR generation HDR visualization (tone mapping)
HDR images examples Image attributions: I, Nattfodd (CC BY-SA 3.0), seng1011 (CC BY-NC-ND 2.0), Abphoto (CC BY 3.0)
Some history... Gustave Le Gray, 1850. Fusion of two negatives.
Study cases 1 Simple case: static scene, static camera 2 Complex case: dynamic scene, hand-held camera
Part I Static scenes / Static camera
Problem Statement Co-registered input images
Problem Statement Co-registered input images
Problem Statement Co-registered input images
Problem Statement Co-registered input images For each pixel position: Input: pixel values HDR generation for exposure times
Problem Statement Co-registered input images For each pixel position: Input: pixel values HDR generation for exposure times Irradiance Map Output: irradiance C number of photos reaching the pixel / unit time
Previous Work irradiance photons/time sensor cell dt =
Previous Work irradiance photons/time sensor cell dt = camera response function pixel value
Previous Work irradiance photons/time sensor cell dt = camera response function pixel value how to get back from?
Previous Work irradiance photons/time sensor cell dt = camera response function pixel value
Previous Work irradiance photons/time sensor cell dt = camera response function pixel value
Previous Work irradiance photons/time combine exposures sensor cell dt = camera response function pixel value
Previous Work Exposure bracketing first proposal Mann & Picard 1995 Different methods propose different weights Gradient based Mann & Picard 1995 Hat function Debevec & Malik 1997 snr based Mitsunaga & Nayar 1999 snr-hat Reinhard et al. 2005 Variance based Robertson et al. 1999, Kirk & Andersen 2006, Granados et al. 2010, Hasinoff et al. 2010
How well can we perform? variance 10 7 10 6 MLE (Granados) Debevec Kirk Mitsunaga Reinhard Robertson Poisson CRLB 10 5 10 4 10 2 10 3 10 4 10 5 irradiance At most 4 samples per pixel! C. Aguerrebere, J. Delon, Y. Gousseau, and P. Musé. Best algorithms for HDR image generation. A study of performance bounds. SIAM Journal on Imaging Sciences.
Part II Dynamic scenes / Hand-held camera
Challenges of HDR imaging in dynamic scenes noise
Challenges of HDR imaging in dynamic scenes noise camera motion
Challenges of HDR imaging in dynamic scenes noise moving objects camera motion
Challenges of HDR imaging in dynamic scenes camera + object motion ghosting effect
Existing methods Treat each problem separately. Camera motion Global alignment adapted to different exposures [Ward 2003] Dynamic scenes De-ghosting techniques [Grosch 2006, Jacobs et al. 2008, Sidibe et al. 2009,Gallo et al. 2009, Heo et al. 2010, Srikantha and Sidibé 2012] Noise Denoising techniques [Buades et al. 2005, Buades et al. 2013, Dabov et al. 2009]
Our approach: simultaneous HDR imaging and denoising C. Aguerrebere, J. Delon, Y. Gousseau, and P. Musé. Simultaneous HDR image reconstruction and denoising for dynamic scenes. International Conference on Computational Photography (ICCP) 2013.
Results: Example 1 Input images
Results: Example 1
Results: Example 1 No ghosting artifacts
Results: Example 1 Reference image Our approach
Results: Example 2 Input images
Results: Example 2
Results: Example 2 No ghosting artifacts
Results: Example 2 Reference image Our approach
Results: Example 3 Input images
Results: Example 3
Results: Example 3 No ghosting artifacts
Results: Example 3 Reference image Our approach
Thanks. Questions?