Camera-Motion and Effective Spatial Resolution

The booming mobile imaging market is aggressively pursuing CMOS imagers with higher spatial resolution and smaller form factors. Given a fixed die size, reducing pixel size seems like a straightforward way to increase spatial resolution. The reduced sensitivity of smaller pixels, however, must be compensated by longer exposure durations. Longer exposure durations result in camera motion blur when the camera is handheld, effectively reducing spatial resolution. We empirically measured how the exposure duration, camera mass and subject’s skill level affect camera motion. We then used these measurements to calculate the effect that camera motion has upon the system MTF for CMOS imagers with different pixel sizes at different scene illuminance levels. We described the scene illuminance conditions for which imaging sensors with larger pixels produce sharper images than imagers with smaller pixel sizes. Introduction Imaging sensor vendors and camera makers have been competing on megapixels and price. The cost of the sensor is most dependent on the size of the die, so designers have crammed more pixels into smaller dice by shrinking the pixel size. Each pixel has a somewhat fixed overhead of space required for circuitry, so as the overall size is reduced, the light sensitive area of each pixel gets smaller [1,2,3,4]. As the light-sensing areas become smaller, the signal-to-noise ratios get smaller as well. To bring the signal-to-noise ratio up to an acceptable level requires a longer exposure [5]. This presents a problem for hand-held photography, since a long exposure will result in motion blur that limits the resolution of the photograph. If the loss in resolution due to motion is greater than the pixel sampling resolution, a higher resolution image could have been obtained with larger pixels and a shorter exposure. This paper considers the tradeoffs between exposure and pixel subtense. Film photographers have long used a rule-of-thumb that a hand held 35mm camera should have an exposure in seconds that is not longer than the inverse of the focal length in millimeters. E.g. that a 50mm lens should have an exposure of 1/50sec at most. This is a very rough approximation at best, and it does not apply to digital photographs. Further, hand shake varies with camera mass and photographer. There have been very few studies of actual hand-held camera shake, so the rule-of-thumb has been difficult to formalize and generalize. In this paper, we carefully measure camera motion as a function of exposure time. We also investigate how camera motion is affected by different photographers and camera masses. We use the measurements to calculate the effect that camera motion has on the system MTF for CMOS imagers with different pixel sizes at scene illuminance levels. We also discuss the applications to antishake system designs. Experimental Setup To measure the effects of exposure duration, camera mass and user’s photography skill level on the amount of camera-shake, each subject was asked to take five pictures of a point light source (a single LED was imaged to a very small spot with a lens, and this spot was placed 9 meters from the test cameras) while holding the camera naturally at each exposure duration (twenty-one total half-stop apart from 0.01 second to 1 second). Seven unpaid subjects (five males and two females, ages from 25 to 50 years old) participated in the experiment. All of the subjects had used digital cameras prior to the experiment. Two of the subjects had used digital cameras extensively prior to the experiment, and they are regarded as “expert users” in the following data analysis. Three cameras were used in these studies (the high-mass Nikon D70, the medium-mass Canon G3 and the low-mass Canon A95) and Table 1 lists the camera-specific parameters. The room lights were turned off during the experiment. To maximize the spatial accuracy of the measurement, all cameras were set at their maximum optical zoom position (4x Zoom for both D70 and G3 and 3x Zoom for A95). Each camera was manual focused on the light source prior to the experiment. In addition, both D70 and G3 were set to output raw image data format to minimize the effect of other in-camera image processing functions. Table 1: Camera parameters Cameras Nikon D70 Canon G3 Canon A95 Mass (grams) 1251 614 335 Degree/pixel 0.00644 0.0063 0.0068 Optical zoom 4X 4X 3X Total pixels 3008x200