Success Stories

Pattern Matching

Results: 21-second FFTs on 40-Megapixel Images (not possible on serial desktop)

The Challenge

Pattern matching is used in a broad array of image processing and security applications: medical imaging, aerospace and defense, security, machine vision, and others. This application has the characteristics that it can be very computationally intensive, it requires interactivity during the algorithm development stage, and it can involve very large data sets.

The application commonly makes use of the Fourier Transform and correlation to locate the target image within the original image. In the increasing number of real-life cases, this type of algorithm needs to operate on very large, high-resolution images, potentially in real time. Also, it is critical to optimize parameters of the algorithm, such as filters and thresholds, to achieve best results.

Star-P Solution

Star-P is well-suited for this application, supporting huge images, rapid processing, and iterative optimization, while still using the familiar MATLAB® environment. Star-P enables interactive optimization of evolving pattern matching algorithms, and enables simple conversion of pre-existing serial code for parallel computing, deals effectively with supercomputing-scale data sets, and delivers supercomputing-class performance.

Summary & Metrics

• Interactive development of image processing algorithms in a familiar MATLAB® environment
• Scale to large and growing data sets, in the gigabyte/terabyte range, for rapid development and real-time processing
• An FFT-based pattern matching application on a 6400x6400 image (1GB of data) ran in 21 seconds that previously could not be run on a desktop PC