Parallel and SIMD Optimization of Image Feature Extraction

Image feature extraction is widely used in content-based image retrieval(CBIR), computer version and all kinds of image processing applications. In this paper, we introduce some parallel and SIMD optimizations of image-feature extraction to overcome the disadvantages of original methods. We mainly use both thread-parallel optimization and Single Instruction Multiple Data (SIMD) optimizations. And especially, for some hot point, we use SIMD logical operations to eliminate the random conditional branches which cannot be effectively predicted by CPU Branch Prediction. We experimented our optimized implementation on multi-core systems, and various images were used to test the image-feature extraction results and performance. All the parallel and SIMD optimizations work out a good cumulative performance speedup.