SPIE Digital Library Get updates from SPIE Newsroom
  • Newsroom Home
  • Astronomy
  • Biomedical Optics & Medical Imaging
  • Defense & Security
  • Electronic Imaging & Signal Processing
  • Illumination & Displays
  • Lasers & Sources
  • Micro/Nano Lithography
  • Nanotechnology
  • Optical Design & Engineering
  • Optoelectronics & Communications
  • Remote Sensing
  • Sensing & Measurement
  • Solar & Alternative Energy
  • Sign up for Newsroom E-Alerts
  • Information for:
    Advertisers


SPIE Photonics West 2017 | Register Today

SPIE Defense + Commercial Sensing 2017 | Call for Papers

Get Down (loaded) - SPIE Journals OPEN ACCESS

SPIE PRESS




Print PageEmail PageView PDF

Electronic Imaging & Signal Processing

Volume holograms for optical correlation

A novel scanning-image volume-holographic correlator enables high-speed optical-pattern matching and tracking.
24 February 2009, SPIE Newsroom. DOI: 10.1117/2.1200902.1504

Pattern recognition of objects within highly cluttered images is a demanding task, although it has many applications in both the military and civilian sectors. The techniques are extremely computationally intensive, making real-time processing using standard computer equipment difficult if not impossible. One possible solution to overcome the computational limitations employs a coherent optical correlator, i.e., an optical processor that allows objects to be identified and tracked at the speed of light.

The coherent optical correlator has had a long and successful history, starting from Vander Lugt's original design. The basic arrangement is still in use today: an input signal or image is coherently encoded on a beam of light, it is subsequently Fourier transformed, and correlated with a holographically stored Fourier signal or filter. The correlated output signal is again Fourier transformed and a peak is produced in the output plane when the filter matches. One of the nice features of this approach is that it is shift invariant, implying that the peak position corresponds to the previously unknown position of the input object. This means that we can identify the object and track its position in a 2D plane.

In modern systems, the filter is implemented on a phase-only spatial-light modulator (SLM) that can be dynamically updated, thus allowing correlation of many templates for real-time target identification. However, the overall bandwidth is limited to the data-bus speed, the frame rate of the SLM, or the frame rate of the camera used for capturing the output. This is a significant weakness.

If one wanted to test 36 filter templates for every video frame, each correlation must take less than 926μs or occur at a frequency higher than 1080Hz. If the filter consisted of 1024×1024 pixels, one would need a data bus of over one gigahertz. If the number of templates were increased further, the speed would become even more problematic.


Figure 1. Schematic layout of the correlator. MQW SLM: Multiple-quantum-well-based spatial-light modulator.

Past attempts to overcome this limitation include our digital/optical hybrid correlator.1 An alternative solution instead encodes the filters within angle-multiplexed volume holograms. The latter have been shown capable of storing tens of thousands of holograms.2 Unlike most holograms, they can achieve a vast amount of storage by employing hologram encoding throughout the volume of the material instead of just on the surface. Their ability to store so many filters removes the need to update a filter SLM. The volume hologram can instead be preloaded with all possible templates. The output is a peak (as in a normal correlator), but its angle or position on the sensor depends on the filter that it matches. We can therefore determine which of the filters the input matches by measuring the position of the peak. And, the shift in angle can be measured easily using a high-speed linear CCD array.

The downside to this arrangement is the loss of spatial invariance, so the correlator will no longer provide the position of the target. We proposed a method to overcome this problem by scanning the input.3 In our architecture (see Figure 1), a video camera of normal resolution (e.g., 512×512 pixels) and frame rate (e.g., 30Hz) acts as the input to the system. The data is held in a high-speed video buffer, and from this a smaller window is extracted (e.g., 64×64 pixels). The window is loaded onto the input SLM, correlated against the volume-hologram filters, and matches are located on the linear CCD array. To restore spatial invariance, the window is moved across the entire input frame one pixel at a time and a correlation map can then be constructed for the full image. The position of the peak indicates which filter has been matched, and the timing gives its location in the input image.

Using this setup, we have removed several of the bandwidth issues associated with traditional correlators. First, the output device is a linear CCD so there is no need to process a 2D image to locate the peak location. A possible megapixel CCD worth of data to process and bus has been reduced to about one thousand pixels. Second, there is now no requirement to update the filter. To scan a 30Hz 512×512-pixel image will require the SLM to work at 409,600Hz. This is achievable using a multiple-quantum-well-based SLM.

We have thus presented a concept for realizing a high-speed optical correlator. The practical bandwidth problems associated with updating the filter and reading out the correlation peak have both been removed by using a volume hologram to store all possible templates and a high-speed linear CCD array at the output. We are continuing our development efforts to further enhance the system's data flow and performance.


Phillip Birch, Rupert Young, Chris Chatwin 
Department of Engineering and Design
University of Sussex
Brighton, UK

The authors are members of the laser and photonics research group.