Research On The Infrared Scene Based Nonuniformity Correction Algorithm

In recent years, with the continuous expansion of application requirements, such as night surveillance, forest fire monitoring, marine search and rescue, marine pollution moni-toring, military targets detection and so on, infrared sensors with high-resolution and high-temperature sensitivity are required. Then advanced remote sensing tasks which are hard for the traditional remote sensors can be completed. However, a problem facing infrared imaging is that the responsivities of the sensor will vary from detector to detector. This results in the nonuniformity and the fixed pattern noises. The thermal sensitivity can be im-proved by reduce the fixed pattern noise or nonuniformity correction. Our research indicates that once the nonuniformity is reduced by0.013%, the thermal sensitivity increases10mk. Meanwhile, as the external conditions change, such as environmental temperature and bias voltage, the responsivities of the detectors drift with time lapse. This increases the difficulty of nonuniformity correction.To reduce the time-drifting fixed pattern noises, many scene-based nonuniformity cor-rection algorithms have been proposed. They can adaptively update the nonuniformity correction parameters to overcome the time drifting responsivities. However,since their high complexity and poor robustness, few matured scene-based nonuniformity correction methods are applied in practical infrared imaging devices. In this dissertation, scene based nonuniformity correction are studied for both staring FPA and scanning infrared imaging system. Our first priority is to improve the robustness of the scene-based nonuniformity correction algorithms. The computation feasibility is also taken into consideration. These researches provides effective theoretical support for the development of high sensitivity in-frared thermal imaging.In the staring FPA infrared imaging system,"ghosting" artifacts tends to occur in IR image after corrected by the existing scene-based nonuniformity correction methods because of their poor robustness. In this dissertation, starting from the infrared thermal conduction effect and analyzing the local spatial correlation of infrared image, the author proposed a scene-based nonuniformity correction algorithm based on temporal and spatial threshold. Decision of updating correction coefficient is made by detecting local spatial correlation and temporal motion, thus preventing edge texture information contaminating correction coefficient and improving algorithm’s robustness. Experiments shows that compared to radiation calibration correction algorithm, the proposed algorithm is capable of reducing residual nonuniformity from0.09%to0.057%.In the scanning FPA infrared imaging system, a nonuniformity correction algorithm based on least squares fitting is proposed. It can reduce the nonuniformity through the infrared sequence. Generalized equations of nonuniformity correction are constructed for each row based on the estimation of real infrared radiation, and the direction of nonunifor-mity correction is adaptively adjusted according to the solution of equations. By utilizing the constraint that the nonuniformity of the IR detectors along the scanning direction can be compensated by the same correction coefficients.Compared to existing SBNUC meth-ods, the proposed one can not only be applied to dynamic scenes, but also to static scene and global static scene with local motion. Experiments suggests that this algorithm is with excellent robustness, and the residual nonuniformity decreases15%-20%. Besides, this dissertation gives the interpretation of the inverse problem of nonuniformity correction of line scanning infrared imaging system. A stripe nonuniformity correction method within single frame of image based on spatial correlation detection is proposed. This algorithm is implemented by using an improved Landweber iteration to solve the inverse problem to get optimal correction coefficient. Compared to other single frame stripe nonuniformity removal algorithms, the residual nonuniformity decreases8%-23%.