Thoughts on Coded Aperture Imaging

The flutter-shutter work shows that temporal coding can preserve high spatial frequencies lost under conventional motion blur. The corresponding question is whether spatial coding can make a defocus point-spread function broad-band and improve recovery from an out-of-focus image.

Coded-aperture techniques have been used for gamma-ray and X-ray imaging, where ordinary optical lenses are unsuitable. Multiple pinholes gather more signal than a single pinhole, but their overlapping images must be decoded. A structured aperture pattern enables that reconstruction. Related multiplexing ideas also occur in medical imaging.

The historical astronomy setting differs from consumer photography. Benefits for point-source distributions do not necessarily carry over to extended natural scenes, and removing the lens can degrade photographic image quality. The cited work discusses the signal-to-noise tradeoff.

Hiura and Matsuyama [1] placed an aperture with four pinholes in the camera and captured multiple images. Their multi-focus system estimated depth through defocus and recovered a blur-free image.

The aperture shape changes blur characteristics and can be selected to preserve information useful for depth measurement. Coded-aperture imaging with a lens therefore predates the later single-image designs.

The analysis by Dowski and Cathey [2, 5] uses Fisher information and Cramér–Rao bounds to study passive ranging. An aperture suited to depth estimation need not be suited to intensity recovery, and vice versa.

The notes highlight an inherent tradeoff for single-lens, single-image capture. Their cited ranging condition requires zeros of the optical transfer function as misfocus varies, whereas robust deblurring seeks a broad-band response.