This paper deals with the problem of infrared image database generation for ATR assessment purposes. Huge databases are required to have quantitative and objective performance evaluations. We propose a method which superimpose targets and occultants on background under image quality metrics constraints to generate realistic images. We also propose a method to generate target signatures with intrinsic thermal variability based on 3D models plated with real infrared textures.

翻译：本文针对用于自动目标识别性能评估的红外图像数据库生成问题展开研究。为获得定量且客观的性能评估，需要大规模数据库支撑。我们提出一种在图像质量指标约束下，将目标与遮挡物叠加于背景之上以生成逼真图像的方法。同时，我们提出基于覆盖真实红外纹理的三维模型，生成具有固有热变异性的目标特征的方法。