The authors of this article propose a technique to compress video signals in the spatial and temporal domains without increasing the system volume or power requirement. They demonstrate the reconstruction of high-resolution video frames from a single coded snapshot image recorded with this method and show various 14 frame and 148 frame videos from everyday examples.
Compressive sensing is a new sampling theory that is based on the sparsity or compressibility of a signal in some transform domain, allowing the entire signal to be determined from relatively few measurements. One way to use compressive measurement for imaging can be implemented by using spatial light modulators to code pixel values incident on a single detector. According to the authors, this strategy increases rather than decreases the bandwidth and operating power due to an increase in data load during the encoding of the signal, as compared to a decreased data load of a compressively sampled signal.
In CACTI, the authors use a harmonically driven binary coded aperture during the exposure of a video capture. In doing so, each temporal plane in the video stream gets modulated by a shifted version of the code, thereby achieving per-pixel modulation at no additional sensor bandwidth. Decoding of the signal is done as in CDMA (code division multiple access) technology. Using an iterative algorithm, temporal channels are separated from the compressed data and several high-speed video frames are reconstructed from a single coded measurement.
CACTI builds on CASSI (coded aperture snapshot spectral imaging), an earlier technique by the same team, in which compressive imaging is used at each plane in a spectral datacube that is modulated by a shifted code. Spectral planes are separated through a dispersive element after coded aperture modulation and detection integrates the spectral planes. The datacube can be recovered by isolating each spectral plane based on its local code structure.
A prototype coded aperture compressive temporal imaging camera consists of a 50mm camera lens, a patterned coded aperture mounted on a piezoelectric stage, a relay lens, and a CCD camera to record the movie. The relay lens projects the spatiotemporally modulated images onto the CCD that records the 30fps coded snapshots. Required high-resolution video frames from a discrete scene are reconstructed from each coded image by using a custom algorithm called GAP (Generalized Alternating Projections).
By using a conventional video camera to record and compressively code the images, the authors were able to reconstruct the videos at 148 frames from a single recorded snapshot. All this was possible with the GAP algorithm that uses sparse representations of a recorded signal. CACTI’s simple coding method might be useful in the future for high-dimensional systems that perform high-resolution spatiotemporal and spectral imaging.
You must log in to add comments.