Image Forgery Detection Using Histogram-Oriented Gradients (HOG)

Due to their capacity to influence people and their usage in news, courts, medical, and military applications, verifying the authenticity of any shared images has become crucial. Image manipulation has become so simple with modern technology. This research introduces a forgery detection method that utilizes the Histogram-Oriented Gradients (HOG) descriptor. This descriptor is established by accumulating the edge orientations caused by image manipulation by a 1D histogram over an image region. The suggested method extracts HOG features in the YCbCr color space of high-frequency discrete wavelet transform (DWT) sub-bands. Later, these features are fed to a classifier to confirm the authenticity of the image or not. Two free datasets, Casia v1.0 and Casia v2.0, were used to evaluate the proposed method. The performance evaluation showed that the HOG descriptor can be utilized to detect image forgery efficiently, with an accuracy of 91.45% for Casia v1.0 and 89.67% for Casia v2.0. The aforementioned accuracies showed that the HOG descriptor can successfully be used to detect image forgery.