A photo composite detection based on eye specular highlights using pixel-based approach

There are several types of photo manipulation. For example, combining objects from different sources into a single image which is often reffered to as photo composite. Previous work proposed by Johnson and Farid estimated the light source through the reflection specular highlights recorded on the eyes. This paper exploits different approach which is pixel-based, to detect photo composite. Our method contains comparative characteristics of the amount of reflected light using EulerNumber, the reflection area using outer line capturing, and the formation of light between objects recorded in the human eye in an image by comparing the distance between the reflection of light. The experimental result describes that the accuracy reach 98% depicted by ROC curve with Area Under ROC Curve (AUC) reach 0.9304. The results fall into Excellent Classification.