HDRLIB: biblioteka do szybkiego przetwarzania obrazów HDR wykorzystująca zaawansowane możliwości procesorów CPU

Development of high dynamic range imaging (HDRI), especially in the display and camera technology, have a significant impact on broader usage of HDR data in image processing, analysis and synthesis. Typical 8-bit graphics pipeline (each red, green, and blue color channels stored in 8-bits) is replaced with the HDR pipeline, in which color is represented by floating point values. The HDRI pipeline does not suffer from many problems of 8-bit systems The most important advantage of HDRI is the possibility of storing full range of luminance and chrominance visible by HVS (Human Visual System). Unfortunately, HDR images are significantly greater than in 8-bit systems. Therefore speed and effectiveness of their processing are especially important for practical applications. In the paper we propose using SIMD and multi-threading CPU technologies to speed-up HDRI processing. We present new architecture of HDRI, in which the attention was paid on memory access optimization, effectiveness of vector data processing and parallelization of computations. We test proposed solution based on a novel and original implementation and discusse achieved speed-ups.