目前藏密分析技術之發展，仍以人為選取藏密特徵為主。但以人為方式選取適當的藏密特徵，並不是件容易的事，同時也可能因為所採用的藏密演算法不同或更動，造成所選取的藏密特徵不再適用，而無法有效偵測。因此，考量人為選取之藏密特徵無法針對個別影像進行調整，本研究探討由演算法自動選取適當特徵並透過機器學習進行藏密偵測之可行性。目標在於利用機器學習將特徵選取及分類整合至同一個架構，並自動從大量的特徵數據中進行訓練與學習。依據此目標，提出了一套植基於倒傳遞類神經網路機器學習架構之空間域數位影像藏密分析方法，能有效偵測以空間域為基礎的藏密技術。並透過繁複實驗測試，證明本分析方法可達到更高之偵測正確率（96.6%），值得將其應用於資訊安全領域中。

The development of steganalysis techniques nowadays is mainly based on manual stego feature selection. However, selecting appropriate stego features manually is not an easy task. Besides, it is also possible that the selected stego features are no longer applicable because the adopted steganography algorithms have been different or modified. In such a situation, effective detection is not guaranteed. Therefore, considering the fact that the manually-selected stego features are not capable of being adjusted to fit the condition of each individual image, this research investigated the feasibility of utilizing an algorithm to automatically select appropriate features and applying the machine learning technique to performing stego detection. The main objective is to use the machine learning technique to integrate feature collection and classification in the same framework, and to automatically complete the training and learning processes among a massive amount of feature data. To achieve such an objective, this research proposed a spatial domain digital image steganalysis (SDDIS) method base on a machine learning technique, namely, the back propagation neural network, to effectively detect the spatial domain-based stego techniques. Various experimental tests have verified that the proposed method can achieve a higher detection accuracy (96.6%) and is worth applying to the field of information security.