A Deep learning model for image retargetting level detection

Detection of image retargeting based on seam carving which is one of the widely used content aware image-resizing algorithm is of importance in image forensics. In addition to machine learning based methods, researchers have developed many other deep learning based methods which involve a binary classifier to determine whether the image is seam carved or not. In this study, we developed a deep learning model that performs multiclass classification for the detection image retargetting level. The proposed model is a multiclass classifier that has four outputs, no (0%) seam carving, 10% seam carved, 20% seam carved and 40% seam carved classes. Training and validation image sets were obtained from ImageNet database. Seam carved images that were reduced by 10%, 20%, and 40% in the vertical dimensions were obtained from these sets. Accuracy analysis of the model was performed by testing model on the images and showing the results of confusion, precision, recall matrices in detail as well as the accuracy value.