一種基於LDPCA之加密灰階影像高效能無失真壓縮機制

Abstract

加密影像的無失真壓縮可以透過 Slepian-Wolf 編碼來達成，而壓縮率會與解碼時如何有效利用資料的相依性有密切的相關。這篇論文為了提高壓縮率，我們從先前已解碼的子影像來預估當前欲解碼子影像上的統計資訊，並利用已解碼的位元平面做進一步精準化。此外，我們提出並實現了一種基於 LDPCA 編碼的高效能無失真壓縮方法。由於LDPCA 解碼的運算程序極其複雜，且為整個系統中最耗時的部分，因此我們使用 CUDA 架構，針對 LDPCA 中的 sum-product 演算法提出一個平行化的設計。為避免運算資源浪費在不必要的運算上，本論文另外提出收斂偵測機制。實驗結果顯示，與先前使用Slepian-Wolf編碼的無損壓縮方案相比，壓縮率可以提高百分之七，且平行化的 LDPCA 解碼器的解碼執行時間比循序的LDPCA解碼器快了約40倍。

Lossless compression of encrypted images can be achieved through Slepian-Wolf (SW) coding, and the compression performance is highly related to how data dependency is exploited while decoding. In this thesis, to improve the compression performance, the statistics of current decoded subimage is estimated from the previous decoded subimages in the same resolution level and then is further refined by the decoded bit planes. Besides, an efficient approach for lossless compressing encrypted images, on the basis of the low-density parity-check accumulate (LDPCA) codes, is proposed and realized. Due to the intricate procedures, LDPCA decoding is the most time-consuming task in our scheme. As a result, a parallelized sum-product algorithm for LDPCA decoding based on CUDA is designed, and an early jump out detection mechanism is also proposed to avoid wasting computational resources on unnecessary operations. Experiment results show that the compression performance is improved about 7% in average, as compared with the state-of-the-art lossless compression scheme using SW coding, and the decoding time using parallel LDPCA decoder is about 40 times faster than the sequential LDPCA decoder.