Digital camera identification based on canonical correlation analysis

Abstract

In this paper, we introduce a new method for digital camera identification from its color images using image sensor noise. We first compute the two noise reference patterns by averaging the noise component from two groups of color images taken with a camera. Then we use canonical correlation analysis (CCA) to calculate the projection directions of the two noise reference patterns. Finally, we calculate the correlation coefficient between the projection of the noise from a specific color image onto one projection direction and the projection of one of noise reference patterns onto another projection direction, then use this coefficient to decide whether the specific color image was taken by the camera or not. Experimental results show that the presented method provides higher accuracy than other methods on the condition of using a few images to compute reference pattern.