The design of medical image transfer function using multi-feature fusion and improved k-means clustering

Wang Xiaopeng, He Shihe, Yu Hu

Abstract

The high quality rendering of important information in volume data is the key to medical visualization. For the purpose to improve unsatisfactory rendering result of the single high-dimensional feature transfer function, a method for transfer function design using multi-feature fusion is proposed. Firstly, the weighed multi-scale morphological opening-closing filter is utilized to remove the noises and small details, and then the features of 3-D volume data such as gray value, gradient amplitude and curvature are extracted. Finally, the improved weighted k-means clustering is applied to construct the transfer function. Since the classification of volume data integrates into multiple features, it reveals more internal structure relations. Medical visualization experiments show that this method increases the contrast of different tissues and obtains better volume rendering quality and efficiency.