| Scaling down traditional optical imaging systems to enhance their form factor presents fundamental challenges in terms of loss of resolution and a decrease in the optical SNR due to the reduced light gathering ability of such scaled imagers. Computational imaging systems can address these issues through joint optimization of their optics and signal processing subsystems. One class of computational imagers is the thin, flat-profile multiplexed imaging system, which uses a combination of several scaled individual imaging units to capture a number of low-resolution images that are then digitally processed to reconstruct a high resolution version of the observed scene.; The performance of multiplexed imagers may be enhanced through the use of adaptive techniques wherein imager resource utilization is maximized through intelligent resource allocation based on the information content in the scene. The body of work laid out in this thesis describes techniques to find regions of interest in a scene and serves to enhance the efficiency of resource allocation in adaptive multiplexed imaging systems. The power spectral density (PSD) is used to derive local entropy maps of input scenes towards identification of regions of interest. Empirical evidence supporting the superiority of PSD-based saliency maps over their histogram-based counterparts in terms of relative local saliency representation of various regions within a scene is provided. Statistical analysis of noise in a scientific-grade digital camera shows that the noise power in images increased with pixel intensity indicating Poisson noise characteristics. A numerically fast and efficient method for computing model-based local saliency maps is presented, and its performance is evaluated in terms of the number of adders, multipliers and table lookups. |
