Region-of-Interest Estimation for Adaptive Resource Allocation in Multi-Aperture Imaging Systems

Abstract

Successful design of a !at-pro"le multiplexed optical imaging system requires the use of adaptive techniques to make intelligent resource allocation based on the information content in the imaging systemDs "eld-of-view. This paper explores techniques for "nding regions of interest in aerial images using local entropy as a descriptor. A novel method for identifying regions-of-interest in images is developed using the 2D normalized power spectral density within GillesD saliency map estimator. Application of the method to candidate aerial images shows its ability to identify consistent regions of interest for such data for varying block sizes and under additive noise. !"#$% &$'()-computational imaging systems, information theory, saliency, image reconstruction

Model-based region-of-interest estimation for adaptive resource allocation in multi-aperture imaging systems

“…Thus, once the parameter α is estimated using (4), the entropy, H(α) as a function of α can be calculated using (6). When k in (6) represents discrete spatial frequency, p k is the discrete PSD, and H(α) in (6) becomes the saliency J(α).…”

“…The normalized PSD p k is a spatial frequency histogram derived from the two-dimensional local PSD. Using a radial spectral model is appropriate for our aerial imaging applications, as the local PSDs of such images exhibit an isotropic nature; thus, we obtain the onedimensional PSD by approximate radial averaging of the 2D PSD [4]. The second term C(α) in (8) is dependent on α, and its value is typically much smaller than that of the !rst term.…”

“…Experimental results in [6] indicated that (1) tends to identify information-rich portions of the scene more readily and consistently than traditional entropy maps using histograms. The main drawback to (1) is its high computational complexity due to the number of terms and nonlinear log evaluations per saliency map pixel.…”

“…Recent work in computational imaging systems has yielded system designs that combine multi-aperature image collection with advanced image reconstruction algorithms in a lightweight "at form factor device [1,2,3,4]. In these systems, multiaperture optical imaging systems built using state-of-the-art micro-optics technology are used for data collection, and multiple low-resolution images are combined using signal processing techniques to produce a !nal image with both a high angular resolution and a large !eld-of-view.…”

“…In [6], a novel approach for generating saliency images or maps was proposed that uses local estimates of the image power spectral density (PSD) to generate the saliency image as…”