Peter N. Crabtree (AFRL), Phan D. Dao (AFRL), Richard H. Picard (AFRL)
Keywords:
Abstract:
Various image de-aliasing algorithms and techniques have been developed to improve the resolution of sensor-aliased images captured with an under sampled point spread function. In the literature these types of algorithms are sometimes included under the broad umbrella of superresolution. Image restoration is a more appropriate categorization for this work because we aim to restore image resolution lost due to sensor aliasing, but only up to the limit imposed by diffraction. Specifically, the work presented here is focused on image de-aliasing using microscanning. Much of the previous work in this area demonstrates improvement by using simulated imagery, or using imagery obtained where the sub pixel shifts are unknown and must be estimated. This paper takes an experimental approach to investigate performance for both the visible and long-wave infrared (LWIR) regions. Two linear translation stages are used to provide two-axis camera control via RS-232 interface. The translation stages use stepper motors, but also include a microstepping capability which allows discrete steps of approximately 0.1 microns. However, there are several types of position error associated with these devices. Therefore, the microstepping error is investigated and partially quantified prior to performing microscan image capture and processing. We also consider the impact of less than 100% fill factor on algorithm performance.
For the visible region we use a CMOS camera and a resolution target to generate a contrast transfer function (CTF) for both the raw and microscanned images. This allows modulation transfer function (MTF) estimation, which gives a more complete and quantitative description of performance as opposed to simply estimating the limiting resolution and/or visual inspection. The difference between the MTF curves for the raw and microscanned images will be explored as a means to describe performance as a function of spatial frequency. Finally, our goal is to also demonstrate algorithm performance in the LWIR region using a microbolometer camera. Current state-of-the-art microbolometers have a pixel pitch of approximately 25 microns, and therefore de-aliasing algorithms are of particular interest for these cameras. Although microbolometers do not provide the best performance in terms of detectivity and are limited to video frame rates, these sensors are important for some applications because they are compact and can operate uncooled.
Date of Conference: September 1-4. 2009
Track: