An Analysis of Image Splicing Detection based on Sensor Noise Levels using Principal Component Analysis (PCA) Estimation against ISO Variation Settings

Image splicing is a process where parts from two or more images are combined to create a single composite image. Since different devices may use varying ISO settings, detecting image splicing could involve analysing the combination of elements from multiple images. This research investigates the effectiveness of using sensor noise levels to detect image splicing, with a particular focus on the influence of different ISO settings. In this research, a comprehensive methodology is used to examine how different ISO settings affect splice detection performance. An image splice detection method is implemented to investigate robustness to different ISO settings in two datasets. The method calculates an estimate of sensor noise based on Principal component Analysis (PCA). The values are then weighted to locate the spliced area. Two datasets are collected, and each dataset shows the combination of two ISO settings images. Experimentation and analysis have revealed a clear difference in detection performance between ISO 800 with ISO 200 and ISO 400 with ISO 320 settings. The results show that the average F-measure score for ISO 400 spliced to ISO 320 is 93.64% which is higher than ISO 800 spliced to ISO 200, which is 87.92%. These mean that lower ISO settings result produce more accurate splicing detection, likely due to lower noise levels in the captured images. As the scores for ISO 400 with ISO 320 are more stable and higher, this stability implies that the method performs consistently well under lower ISO settings, making it more reliable for practical applications.