کد مقاله | کد نشریه | سال انتشار | مقاله انگلیسی | نسخه تمام متن |
---|---|---|---|---|
494463 | 862796 | 2016 | 11 صفحه PDF | دانلود رایگان |
The fast continuous rotation scanning mode is suitable for situations when a higher scanning speed or a lower radiation dose is required for micro-CT. Under this scanning mode, the specimen rotates far beyond 2π, but not an integral multiple of 2π continuously while the detector collects projection images at a high frame frequency, which consequently produces uneven-distributed and redundant projections with a high level of noise. Therefore, for subsequent three-dimensional (3D) image reconstruction, each 2π-projection view during a whole rotation span must be accurately located, and image denoising is an essential pre-processing step for improving quality of reconstructed images. For matching the rotation angle with the projection image accurately under the fast continuous rotation scanning mode, a structure similarity (SSIM) coefficient was used as a control parameter for extraction of periodic projection sequences for 3D reconstruction. Meanwhile, an improved non-local means (NL-means) algorithm was proposed for noise reduction in projection sequences, and a graphic processing unit (GPU) capable of highly parallel and fast floating-point calculation was used for alleviating the computation cost of the algorithm. The experimental results show that the SSIM-based 2π-projection-view locating method is highly accurate in identifying periodic projection sequences and is easy to implement, and that the improved NL-means algorithm can well restrain the quantum noise while preserving detailed information.
Journal: Neurocomputing - Volume 207, 26 September 2016, Pages 335–345