Image Restoration In High-resolution Panchromatic Remote Sensing Images Processing

Interpretation and mapping from panchromatic high-resolution remote sensing images is very important for military applications. In the military applications, structure information is the most important for special analysis and judgement. But unfortunately, there is always unavoidable image degradation caused by many factors. These degradation would come from the detectors, the recorders, the atmosphere, the noise etc. The image degradation would introduce some uncertainty in the military analysis and judgement. Image restoration techniques are oriented towards modelling the degradations, blur and noise, and applying an inverse procedure to obtain an approximation of the original scene. It is a basic technique for image analysis and object recognition.In this thesis, two image restoration problems in panchromatic remote sensing images processing are considered: image deconvolution and haze removal.For the image deconvolution problem, a procedure of estimation of point spread function (PSF) over linear ground edge is first discussed. Error analysis is also implemented through numerical simulation experiment. After a thorough analysis of the ARTUR algorithm which has the virtue of edge preserving property, a modified fast algorithm based on biorthonormal wavelet is proposed. Compared with the original ARTUR, this algorithm could achieve almost identical results with less iteration number. As remote sensing images are generally very large, it's generally impossible to process the whole image in the memory. An overlapped block processing method is proposed to handle this situation and reduce the whole processing time.For haze removal, a new automated detection and removal procedure is proposed based on the undecimated wavelet transform. Compared with homo-morphic filtering technique, this proposed algorithm could estimate the optical thickness of the hazy image regions and automatically adopt different processing amplitude. It could achieve better results and keep the haze-free region remain invariable. Processing parameters of the algorithm are also easier to be deter-...