Abstract: In order to provide the specific information of location and shape of lumbar discs for the computer-aided diagnosis of lumbar disease, an algorithm of lumbar discs localization and segmentation based on CT images is proposed in this paper.In consideration of the disturbance of skeleton and complicated textures of the background, local features are extracted using wavelet decomposition, and the high-dimensional feature space is transformed into assemble of multiple one-dimensional feature spaces by introducing independent component analysis, apre-selection algorithm for lumbar discs based on visual saliency detection is presented to realize the evaluation for the possibility density of the wavelet features and the saliency map of lumbar discs is obtained.Besides, the local samples are randomly extracted based on both the saliency map and the orientation information measurement, and a new category of weak classifiers for sparse features is designed to improve the recognition rate of the Hough forest algorithm and to realize the accurate localization and segmentation of lumbar discs.The experimental results show that the proposed algorithm is effective in improving the accuracy and the computational speed of lumbar discs localization and segmentation.