부산대학교 도서관

JPEG fragment carving technology is a hot and difficult research field in data recovery and computer forensics. Most of the adjacent Pixel Similarity algorithms for JPEG images have high accuracy when the pixels are continuous and smooth, Low accuracy will occur when image contains sharp areas such as there are strips, vertical strips and angles. SoD and ED only focus on the similarity between adjacent edge pixels, rather than the integrity and smoothness of the whole image. MED uses the surrounding pixel values to predict the target pixel value, which is better than SoD and ED in the prediction of pixel integrity. In order to solve this problem, we provide a new Euclidean distance adjacent pixel detection algorithm MED-ED based on median edge detector, according to the local stability and the characteristics of MED. Secondly, the design algorithm to judge the original cluster size, according to the Features of the file system, in order to solve the problem of inefficiency under the situation of the current storage device capacity soaring and high image quality. The experimental results show that the proposed method can effectively improve the accuracy and efficiency.