Research On Enhancement Algorithm For Low SNR IR Target Images

This dissertation has analyzed the imaging principle of Infrared Focal Plane Array(IRFPA) thermal imaging system, primary noises in the IRFPA imaging system and primary features of infrared images. Infrared images are created by temperature differences between targets and their background. Fixed pattern noises, noises generated by blind pixels and random noises are primary noises of IRFPA thermal imaging system. Great bright background and low contrast between targets and background are the primary characteristics of infrared images. These result the poor image quality and visual impression. Reducing noises and enhancing SNR are two important steps of infrared images processing.Based on the analysis to the noises generating schemes and their characteristics, several different de-noise and enhancement algorithms are proposed to restrain noises influence and enhance SNR of infrared images. For fixed pattern noises, emphases are laid on the research of non-uniformity correction for IRFPA based on neural net. A new non-uniformity correction algorithm based on region feature classification and a new non-uniformity correction algorithm based on infrared image sequences are proposed in this dissertation. These two algorithms can update correction parameters to improve the correction effect according to the scene changes. A new algorithm based on motion analysis to compensate the blind pixels in IRFPA. Motion field of targets is proposed. By making motion analysis to IR image sequences, blind pixels are compensated by use of relativity of neighboring frames. And the edges of targets are preserved well. For the background noises in IR images, an adaptive image enhancement algorithm based on plateau histogram equalization are presented in this dissertation to improve the contrast of IR images. By adaptively selecting plateau threshold value, this algorithm can enhance targets and constrain the background noises. It overcomes the disadvantages of histogram equalization and has small computation amount and is implemented real-timely. At the same time, this dissertation proposed a multi-resolution motion field estimation algorithm based on wavelet transformation for noisy IR image sequences. This algorithm can estimate the motion vectors without the influence of noises. On the base of motion field estimation, image sequences enhancement and de-noise algorithms based on motion vector classified are proposed. The targets in IR images are avoided of being blurred when IR images are de-noised by image sequences noise reduction algorithm. And the IR image sequences are enhanced at the same time that their noises are constrained.At last, an IRFPA real-time processing imaging system is designed according to the characteristic of output signal of IRFPA, signal channel configuration, high frame frequency of IRFPA and the targets'and backgrounds'features of imaging guided tactics missile.128x128 IRFPA real-time signal processing system is implemented to correct the fixed pattern noises, compensate the blind pixels in IRFPA and enhance the contrast of IR images. The imaging effect of IRFPA imaging system is improved greatly. This system has advantages of great dynamic range, rapid operation speed, high operation precision, stabilized performance, high reliability and high agility.