Virtual camera-based analysis of photosensor characterization methods

Abstract

Digital cameras are widely used in artificial intelligent systems: machine vision, pattern recognition, automatic defect detection, etc. The quality of these systems is defined in significant degree by the registered images signal-to-noise ratio. Digital camera noise, conditions of shooting, low-light illumination decrease image quality. But information about camera noise allows to significantly improve it. However, noise characteristics are often absent in the instructions even for scientific digital cameras. There are various methods of photosensor noise measurement. They have different accuracy, speed of implementation and processing time. Therefore, the choice of the optimal method for measuring noise is of great interest. The paper presents the virtual camera-based analysis of methods of photosensor characterization. This technique determines the maximum and possible accuracy of the methods. Three methods of photosensor noise characterization are considered: the EMVA (European machine vision association) standard 1288, automatic segmentation of a non-uniform target (ASNT), and automatic segmentation of a striped target (ASST). The method's accuracy and processing time were estimated.