Image Forgery Detection Using Structural Similarity and Morphological Tools

Recent advancements in digital imaging technology and computer vision have brought forth a multitude of new possibilities in every field ranging from manufacturing to healthcare and graphic designing to virtual reality. With the many benefits that these advancements have brought when this technology falls into unscrupulous and malicious hands its effects can be adverse. Nowadays, it is possible for anyone to easily manipulate images to an extent that makes it seamless to the human eye, making it impossible to distinguish between an original and a counterfeit image. An immensely popular example of this technology is deepfake technology, which has the potential to create entirely false narratives of people as the results of deepfake media have started to look stunningly believable. To be able to distinguish what is real and what is not, we need to be able to measure the structural similarity between both real and fake images. For this purpose, we need a metric that can effectively distinguish different structures in an image. Two such widely used metrics are Mean Squared Error (MSE) and Structural Similarity Index Measure (SSIM). These metrics can also be used in assessing the quality of an image by comparing it with a lossy and compressed version of itself to understand the degradation in quality. This research applies morphological tools over MSE and SSIM and provides a series of experiments. The intent is to compare both metrics and conclude which of the two provides us with a more accurate description of the structural differences between two images. Our result shows that using a combination of morphological tools i.e., Erosion and Dilation (collectively called Opening), with SSIM provides better results to find the differences or similarities among images.