Image Splicing Detection Using Retinex Based Contrast Enhancement and Deep Learning

With the evolution of many user friendly multimedia tools such as adobe Photoshop, manipulating an image has become a facile task. Image splicing is one among various existing forgery techniques where an image is merged with one or more images with the intention of changing the originality of given image. The significance of this article is to evolve a new model to find out the image splicing forgery using deep learning technique. In initial stage the given image is by converted into YCBCR format and by enhancing its contrast using Retinex Model which will enhance the efficiency of the proposed architecture. Then the image characteristics are extracted using the popular ResNet-50 network. Finally, SVM classifier classifies the gained characteristics as spliced or Real. The overall accuracy achieved by the proposed architecture is surpassing the several existing referenced techniques.