An improved painting-based transfer function design approach with CUDA-acceleration
Deqing Qu, Yuetong Luo, Wenmin Tan · 2011
By coupling machine learning and painting metaphor, painting-based transfer function design approach allows more sophisticated classification in intuitive manners. With the aim of improving classification performance for noisy data, statistical properties such as mean value and standard deviation have been used instead of intensity and gradient magnitude to eliminate disturbance of noise. To achieve immediate feedback in painting process, both machine learning method, i.e. Artificial Neutral Network, and volume rendering are implemented by CUDA. Furthermore, the effectiveness of our method has been testified through experiments on both synthetic data and real data with noise.