Replication Image Detection Using Convolutional Neural Network

  • ICIPPSD Conference


The ease of access to image data has led to overuse of repeated images at various instances which leads to increases duplication and redundancy in many industries. Advanced editing techniques which are available very easily, encourages original copyrights images to be misused. This results in lack of originality in data generated at every level. Common solutions include allowing manual selections of duplicate images or compares images pixel by pixel. The conventional method is to use 3 branch Siamese Convolution model to detect duplication of medical images. We propose to develop a model to detect duplication in everyday images by training a Siamese Convolutional Neural Network and try to achieve greater accuracy than previously developed solutions. Using Grad-Cam network inspection we propose to inspect the decisions taken by the CNN upon detecting duplication in images.