Detecting Resized JPEG Images by Analyzing High Frequency Elements in DCT Coefficients

In this paper, we propose a method for detecting resized JPEG images. We defined 8 × 8 periodic blocks as JPEG blocks. JPEG block boundaries are detected by applying 8 × 8 block discrete cosine transform (DCT) to all the pixels of the input image and analyzing the high frequency coefficients in them. In order to quantitatively analyze the degree of forgery, we have developed two approaches such as truth-score and correlation-score methods. Experimental results using 375 original (untouched) JPEG images and 2,250 resized images recompressed with a variety of quality factors demonstrated that our proposed method can classify them with over 90% of accuracy. Our proposed method can detect not only conventional resizing but also state-of-the-art non-linear resizing such as seam carving.