Simd Optimization of Euclidean Distance Transforms for Pattern Recognition

This paper describes a SIMD optimization method for computing different Euclidean distance algorithms. Distance transforms have been widely applied to image analysis and pattern recognition problems. The proposed approach is based on the inherent fine and medium-grain parallelism of considered distance algorithms and has been implemented using Intel Streaming SIMD Extensions (SSE), intrinsics and VTune Analyzer. Experimental results show that optimized prefetched SIMD algorithms improve by four the number of execution cycles in comparison with the initial SISD solutions.