自從911 事件發生後，資訊隱藏及偵測技術成為各國國防安全上的熱門研究課題。 預測差值直方圖位移藏密法(PEHS)可有效的在影像中藏入大量秘密訊息，但其在預測差 值的直方圖上卻有不正常的分佈。本論文提出可有效偵測PEHS 藏密影像之偵密技術， 主要是透過預測差值直方圖的比率來進行特徵值的擷取，並利用倒傳遞類神經網路進行 分類，以分辨是否為使用PEHS 之藏密影像。實驗結果證明，本文所提出之偵測技術對 PEHS 偵測正確率達99%以上，與2007 年Zhang 學者與2010 年Pevný 學者提出之偵測 方法所提供之69.7%與70.9%偵測率比較，可有效提升對PEHS 藏密影像的偵測正確率， 適合國防安全上之應用。

Information hiding and steanalysis have become hot research topics in the field of national defense security after the 911 tragedy. The Prediction-Error Histogram Shift method (PEHS) can effectively hide large amount secret message in an image. However, the histogram of the Prediction-Error differences appears abnormal distribution. In this paper, we propose a steganalysis method which can effectively detect the stego images created by PEHS. The proposed method uses the ratio of Prediction-Error to extract the image feature. The image feature is then classified by Back-Propagating Neural Network. Experimental results show that the accurate detection rate of the proposed steganalysis method to PEHS is above 99% and outperform the performance of 69.7% and 70.9% provided by the method proposed by Zhang et al. in 2007 and Pevný et al. in 2010. Therefore, the proposed steganalysis method can effectively improve the detection rate for detecting the stego images created by PEHS and is practical for applications of national defense security.