Accelerating Iris Template Matching using Commodity Video Graphics Adapters

As iris technology becomes widespread, template database sizes will grow. High resolution iris capture systems such as "Iris on the move" and "Iris at a distance" will require real-time template matching to be utilized in non-static, multiple target environments such as airports or subways. When real-time iris capture systems are interfaced to national-sized databases, higher template matching speeds will be a necessity. A utilization of the immense computational power contained within commodity video graphics adapters is a cost effective means of attaining higher template matching speeds. In this research, a highly optimized C++ iris template matching algorithm is ported to a video graphics card's graphics processing unit using the high level shader language (HLSL) which is part of Microsoft's DirectX 9.0. To demonstrate the acceleration gains, template matching in the form of a Hamming distance calculation is performed on large iris template databases. Results show that a state-of-the-art video card performed template matching 9.5 times faster (24.65 million template comparisons per second) than a CPU. A single computer containing four video cards could approach speeds of 100 million matches per second. A detailed methodology and code sample are presented, and technical issues are discussed. Methods are provided for improving these speeds using new features in DirectX 10.