We present a distance metric learning algorithm for regression problems, which incorporates label information to form a biased distance metric in the process of learning. We use Newton's optimization method to solve an optimization problem for the sake of learning this biased distance metric. Experiments show that this method can find the intrinsic variation trend of data in a regression model by a relative small amount of samples without any prior assumption of the structure or distribution of data. In addition, the test sample data can be projected to this metric by a simple linear transformation and it is easy to be combined with manifold learning algorithms to improve the performance. Experiments are conducted on the FG-NET aging database, the UIUC-IFP-Y aging database, and the CHIL head pose database by Gaussian process regression based on the learned metric, which shows that our method is competitive among the start-of-art.

Most digital still color cameras use a single electronic sensor (CCD or CMOS) overlaid with a color filter array. At each pixel location only one color sample is taken, and the other colors must be interpolated using neighboring samples. This color plane interpolation is known as demosaicking, which is one of the important tasks in a digital camera pipeline. Demosaicked images possess spatially periodic inter-pixel correlation. In this paper, such correlation is expressed in a quadratic form, and Principal Component Analysis is applied to filter out intrinsic scene correlation. A decision mechanism using BP neural networks and a majority-voting scheme is designed to recognize demosaicking correlation and authenticate digital photos. Experiments show that, the proposed method can accurately classify images by demosaicking algorithms or source cameras, and it is effective to detect rendering forgeries. The sensitivity and robustness of the method are also verified. This algorithm-independent approach is especially useful when demosaicking algorithm is only available in form of binary code or integrated circuit without technical detail.

We represent the spatially periodic inter-pixel correlation due to color filter array interpolation in a quadratic form. Based on this, a coefficient matrix is obtained for each color channel, whose principal components are extracted and fed to feed-forward back propagation networks for source camera identification. Experiments demonstrate the efficacy, sensitivity and robustness of our method