Recaptured Image Forensics Based on Image Illumination and Texture Features

Advanced devices like digital cameras and smart devices have triggered the greater possibility of recapturing high-quality images from output media. These high-quality recaptured images are genuine enough to escape the detection of normal human eyes, and the existing image forensic techniques. Giving the malicious users opportunities to distribute these fraudulent images to execute illegal activities. Due to the difference in illumination environments during original scene capture and recapturing processes, the recaptured image exhibits different illumination feature patterns compared with the original scene images. Furthermore, the fact that during every image capturing event the camera compresses the images, implies that a recaptured image suffers double compression which significantly alters its texture features. Various studies have showed illumination and texture features to be key features for generating statistical patterns that can give pertinent information for recaptured image detection. This study proposes an efficient method that combines the illumination and texture features for enhancing the recaptured image detection. The proposed approach begins with dividing the image into 16x16 blocks, extracts texture and illumination features from each image block and finally aggregates the extracted features for classification. Evaluated on a publicly available image dataset with high quality recaptured images, the proposed approach recorded good detection results with a detection accuracy of up to 94.4%, a recall of 95.6% and a false detection rate of 6.7%.