Implementation of New Strategies in Multidimensional Compressive Imaging

[摘要] Since 2004, compressive sensing (CS) has attracted considerable attentions due toits virtue of being able to reconstruct a signal when being sampled at sub-Nyquist rate.Building on the mathematical breakthroughs of CS, we previously developed a uniqueimaging hardware platform, named single pixel camera (SPC), which incorporates aspatial light modulator and a single detector. We have exploited this to construct theinfrared, hyperspectral, and low-light imaging systems that have greatly reduced costin power, space, and/or expense compared to their traditional counterparts. However,previous imaging applications based on these systems are constrained to static imagesand fail for time-varying scenes (videos). There are also several drawbacks to theSPC system. One of which is the use of the digital micromirror device (DMD) asthe light modulator. While this modulator has a broad spectral response from theultraviolet through the visible and across the mid-infrared, it does not operate inthe terahertz or other portions of the spectrum. Additionally, the binary nature andlimited twelve degree of tilt of the micromirrors reduces theeld of view and presentsoptical design challenges in certain con gurations. To surmount the static scenelimitation, we implement, compare and analyze two new algorithms for compressivevideo acquisitions and reconstructions. We also present therst experimental resultsfor video recovery by combining the algorithms with the SPC system. To overcomethe drawbacks of the use of the DMD as the light modulator in the SPC, we proposean implementation of a new system based on a single mask design whose lateraltranslation achieves the same purpose as the DMD during data acquisition but witha simpler, more robust form factor.

[发布机构] Rice University

[效力级别] sensing