?
Алгоритм обратимого сжатия серий медицинских изображений с выборочным устранением межкадровой избыточности
The growing number and resolution of computed tomography examinations increase the load on the digital archives of medical organizations, while lossless compression remains mandatory. The lossless schemes admitted by the medical image interchange standard are intra-frame and ignore the correlation between adjacent slices, whereas known inter-frame methods choose the prediction mode for the whole frame, disregarding how homogeneous and edge regions are distributed within it. The paper proposes a lossless coding algorithm for medical image series that extends the JPEG-LS standard with a stage of selective inter-frame decorrelation of prediction errors. The frame is partitioned into non-overlapping blocks of eight by eight pixels; blocks containing anatomical boundaries are identified by the Canny edge detector, and for each of them a spatially matched block is searched for in the reference (previous) frame using the normalized cross-correlation criterion. For such blocks, the difference between the prediction errors of the current and reference blocks is encoded, which reduces the variance of the encoded value at the input of the entropy coder. An experimental evaluation on eighteen computed tomography series comprising 6213 frames showed a mean gain in compression ratio of 18.44% and 22.81% over JPEG-LS and JPEG2000, respectively.