This video demonstrates the application of the compressed sensing algorithm to an X-ray backscatter imaging system. X-ray backscatter imaging is inherently slow, with full-resolution scans taking over an hour to acquire. By using compressed sensing, a high-fidelity image can be reconstructed from a small, randomly sampled subset of pixels, drastically reducing acquisition time to under 20 minutes. The video details the construction of the scanner, which uses a motorized gimbal, potentiometer-based position sensing, and a custom photon-counting detector built with a photomultiplier tube and a Teensy microcontroller. The core principle of compressed sensing is explained: it leverages the fact that all "natural" images are sparse in the frequency domain, allowing the algorithm to find the most probable reconstruction using the fewest unique frequencies that fit the sampled data.
Leveraging Compressed Sensing for High-Speed X-Ray Imaging
0:00:01 The Magic of Compressed Sensing: The video introduces compressed sensing, an algorithm that can reconstruct a full image from a small fraction (e.g., 10%) of its pixels. This is especially useful for imaging techniques where acquiring each pixel is time-consuming.
0:01:05 X-Ray Backscatter Explained: The system works like "X-ray goggles," scanning a thin beam of X-rays across an object and detecting the reflected (backscattered) photons. Objects with low atomic number elements (like hydrogen, carbon, oxygen) appear bright, while dense, high atomic number elements are dark.
0:02:58 Overcoming the Speed Barrier: A full 800x800 pixel scan would take approximately 85 minutes. By sampling only a fraction of the pixels and using compressed sensing, the acquisition time is reduced to a more practical 10-20 minutes with comparable quality.
0:04:05 Random Sampling is Key: The algorithm requires random, not regular, sampling to avoid blind spots and accurately determine the image's frequency components. The mechanical imprecision of the scanner's movement naturally provides this necessary randomness.
0:08:01 Scanner Construction Details: The scanner is built on a gimbal mount with DC gear motors for movement. Its position is tracked using simple analog potentiometers connected to a Teensy microcontroller, which provides an absolute position easily.
0:09:32 Photon Detection System: The detector consists of a phosphor screen, a photomultiplier tube (PMT), and a transimpedance amplifier. A Teensy microcontroller efficiently counts the resulting voltage pulses using its built-in hardware comparator and counter, freeing up the CPU for other tasks.
0:12:43 How the Algorithm Works: Compressed sensing reconstructs the image by finding a solution that fits the measured data points while using the fewest number of unique frequencies. This is a challenging optimization problem.
0:15:34 The "Natural Image" Principle: The algorithm's success is based on a fundamental property of the universe: all "natural" images and signals (photos, audio, etc.) are sparse in the frequency domain. They occupy a tiny sliver of all possible images, distinguishing them from random "noise" images, which are not sparse.
0:20:13 Open Source Information: All project details, including the code for the control system and reconstruction algorithm, are made available on GitHub for viewers to explore further.