- 學位論文
基於閾值控制的AMBTC壓縮影像之驗證技術
Authentication Technology for AMBTC Compressed Images Based on Threshold Control
摘要
現今數位影像在網路上傳輸時，一般會透過壓縮技術將數位影像轉換成資訊量較小的壓縮碼進行傳輸。但壓縮碼在傳輸的過程中，有可能會遭受到惡意的篡改或破壞，若接收者沒有檢測到就有可能得到錯誤的資訊。因此為了判斷影像是否遭受到篡改，許多學者陸續提出了影像驗證的相關技術。過去曾有學者提出了基於matrix encoding的影像驗證技術，主要作法是將影像的特徵當作驗證碼嵌入到AMBTC壓縮碼的量化值中，最後根據驗證碼的完整度來判斷影像是否遭受到篡改。但是本研究觀察到，在該學者的做法中平滑塊內的像素值越接近，驗證碼嵌入時造成的破壞就越嚴重。且該學者的做法無法偵測到針對AMBTC壓縮碼中，量化值前5個MSB的篡改。因此本研究提出了一個基於閾值控制的技術，將壓縮碼分為平滑塊及複雜塊，平滑塊指的是區塊中像素值較為接近的區域，反之則為複雜塊。最後根據這2種區塊的特性，分別使用不同的嵌入法將驗證碼嵌入到壓縮碼中。在平滑塊中，本研究使用matrix encoding的技術翻轉位圖來完成驗證碼的嵌入。複雜塊的部份驗證碼仍然會透過APPM的技術嵌入到量化值中。在產生驗證碼的過程中，皆有將量化值的前5個MSB加入計算，藉此保護MSB的資訊。實驗結果顯示本論文提升的PSNR值最高達到了3.25 dB以上，最低也提升了0.36 dB。不但可以偵測到針對量化值前5個MSB位元的惡意篡改方式，也成功的得到了較佳的影像品質。
並列摘要
The current digital images transmitted on the internet are generally compressed into smaller sized compressed codes through compression technology. However, during the transmission process, the compressed code might be maliciously tampered with or corrupted. If undetected, the receiver will get the wrong information. Therefore, to determine whether an image has been tampered with, many scholars have proposed techniques for image verification. In the past, some scholars have proposed the image authentication technique based on matrix encoding. The main method is to embed the image features as authentication code into the quantization value of the AMBTC compressed code, and finally determine whether the image has been tampered with based on the condition of the authentication codes. However, it is observed in this study that in the scholar's approach the closer the pixel values from the smooth block the more severe the damage caused by the authentication code embedding. The approach is unable to detect tampering of the first 5 MSBs of the quantized values in the AMBTC compressed codes. Therefore, this study proposed a technology based on threshold control, which divides the compressed codes into smooth blocks and complex blocks. The smooth block refers to the area where the pixel values are relatively closer in the blocks, and vice versa in the complex blocks. Finally, according to the characteristics of these two blocks, different embedding methods are used to embed the authentication codes into the compressed codes. In the smooth blocks, this study uses the technique of matrix encoding to flip the bitmap to complete the embedding of the authentication code. Whilst in the complex blocks, the authentication codes are embedded in the quantized value by APPM technique. The first 5 MSBs of the quantified values are added to the calculation in the process of generating the authentication codes to protect the MSB information. Experimental results showed that the PSNR value is improved by more than 3.25 dB and by at least 0.36 dB. It can also detect the malicious tamper and corrupt for the first 5 MSB bits of the quantized value. The image quality is also successfully improved.