Dictionary learning based denoising of low-dose X-ray CT image.
Résumé
A dictionary learning based denoising method is introduced to eliminate the noise in low-dose computed-tomography(LDCT)image. Aiming at the phantom and patient images, the k-means singular value decomposition(K-SVD)algorithm is adopted to train image dictionary iteratively based on LDCT and normal-dose CT(NDCT)images. Then, based on the orthogonal matching pursuit algorithm, the sparse representation decomposes the noise image into sparse component which is related to image information and remains which are regarded as noise. Finally, noises can be suppressed by reconstructing image only with its sparse components. The experimental results show that the performance of the proposed method is strongly affected by the dictionary size and the constraints for sparsity in dictionary training. The better performance can be achieved when training samples are extracted from NDCT image instead of LDCT image. Under the same noise level, compared with the traditional de-noising methods, the proposed method is more effective in suppressing noise and improving the visual effect while maintaining more diagnostic image details.