Gradient Camera
We propose a gradient-based camera that relies on ratios of neighboring pixel intensities rather than absolute intensity measurements. A gradient camera is similar to existing intensity cameras in electro-optical structure, but by measuring only the local changes in the image, we gain some significant advantages. It is most similar to the locally-adaptive gain cameras. However, unlike locally-adaptive gain cameras where the gain is different for each pixel, in gradient camera the local gain is same for each clique of four neighboring pixels. Since the ratio between neighboring pixels in the clique is invariant to the gain, we do not need to record the gain. Gradient camera needs little or no exposure metering to capture high contrast scenes, hides effects of quantization well and distributes noise as low-frequency error rather than masking high frequencies.
Background & Objective: Capturing images of a scene with large variations in intensity requires a camera with a high dynamic range. This is sometimes achieved with image sensors with logarithmic response curve. On the other hand, successful biological visual systems (including the human eye) use sense change more acutely than absolute intensity. Our goal is to build a gradient camera than captured subtle detail as well as large variations in intensity.
Technical Discussion: Quantizing sensed intensity differences between adjacent pixel values permits an ordinary A/D converter to measure detailed high contrast, high dynamic range (HDR) scenes. Once we have computed pairwise ratios of intensities, we reconstruct the original image using a 2D integration by solving a Poisson equation. The pairwise ratios eliminate effect of unknown gain within each clique. We measure alternating "cliques" of sensors (small groups) that locally determine their own best exposure. This intrinsically differential design suppresses common-mode noise, hides and smoothes quantization, and can correct for its own saturated sensors. Simulations demonstrate these capabilities in side-by-side comparisons.
Outside Collaborations: Jack Tumblin, Northwestern University
Contact: Amit Agrawal
Technology Area: Sensor and Data Systems
Modification Date: July 7, 2008
