Source Camera Identification Using Hybrid Feature Set and Machine Learning Classifiers

Abstract

AbstractSource camera identification (SCI), which is used to identify the source camera of the images/photos, plays a very important role in today’s era especially in the domain of digital image forensics. Earlier, photo response non-uniformity (PRNU)–based methods were used to identify the source camera of images, but it involves a lot of complexity as PRNU noise gets highly contaminated by image scenes which results in the wrong classification. With the growth in machine learning, researchers developed several methods that work on features extraction and then training them on classifiers which results in better results than PRNU-based methods, but due to limitations of feature space (spatial features), the performance decreases when either number of classes increases or properties of images belong to the same class are different. This chapter proposes a framework based on the frequency and spatial features by performing data augmentation to increase the diversity of images in the dataset and extracting a sufficient amount of feature vectors. These features are extracted from images by applying discrete wavelet transform techniques (DWT) and local binary pattern (LBP). Further, these are trained on multi-class classifiers such as support vector machine (SVM), linear discriminant analysis (LDA), and K-nearest neighbor (KNN). The comparative study of experimental results shows that the proposed method outperforms other state-of-art methods.KeywordsSource camera identificationMachine learning classifierDiscrete wavelet transformLocal binary pattern