Automatic classification of noise for infrared images into processes by means of the principal component analysis

Noise characterization and classification is an important task to evaluate the performance of an infrared imaging system. The focal plane array infrared cameras present several types of noises: fixed pattern noise, 1/f noise, pure temporal noise, etc. The existence of bad pixels showing a singular behavior must be included in the noise description. In this paper we show how the principal component analysis is able to classify the noise of a set of frames into different subsets. The classification method is integrated into a software package that performs the classification of the obtained eigenimages into processes. This method is specially adapted to the analysis of noise in a set of frames because it produces a corresponding set of images characterizing the noise. A result of the analysis provided with this method is the extraction of the fixed pattern noise, the bad pixel identification, the 1/f noise components and analysis, the pure temporal noise, and some other processes having intermediate time scales.