Research On Synthetic Simulation Of Optical Remote Sensing Images

The simulation of remote sensing(RS)images are techniques that generate RS simulated images based on the RS theoretical models,prior knowledge and RS image features.By applying such techniques,abundant of RS simulated images with different parameters can be yielded,which are of great significance to the design and evaluation of RS image interpreting algorithms,and then are available in both of the military and civil fields.Recently,the simulation of RS images is mainly performed by simulating the RS imaging procedure.These methods generally refer to lots of complicated imaging taches with many problems such as the difficult mathematical description and the low efficiency.Then,they often cannot meet the need of the high real-time applications.Therefore,it is an urgent problem to improve the efficiency for simulating RS images.Actually,for various specific applications,only the characteristics of RS images are concerned,instead of the specific taches of RS imaging procedure.For this consideration,this paper studies the methods of simulating RS images from the viewpoint of the statistical characteristics of RS image,where the probability distribution and the spatial correlation is considered.A novel scheme of simulating RS is proposed.At first,by some image segmentation methods,the RS images are divided into the background and foreground regions.Then these two parts are simulated respectively.Finally,the background and foreground images are fused with some appropriate strategies,leading to the final simulated RS images.The main work of this paper are as follows.Firstly,a segmentation algorithm of RS images is studied based on the local Simple Linear Iterative Clustering(SLIC)method.To address the weakness of the traditional image segmentation methods based on clustering,including their low efficiency and lacking of using the spatial information,a fast segmentation method for RS images based on local SLIC is proposed.In this method,a fast preprocessing step to determine the region of interesting is introduced,which reduces the image scope for pixels clustering and then improve the algorithm's efficiency.The experimental results have validated the performance of the proposed algorithm.Secondly,the method of simulating RS background image based on random field theory.The RS background image is assumed to be two-dimension random field with a specified probability distribution and spatial correlation,and then,an algorithm of simulating RS background image is proposed by combining the non-linear transform and spectrum methods.The experiments show that this algorithm can generate realistic simulated RS images.Finally,an algorithm of fusing the RS foreground and background images is studied.The problem of fusing the foreground and background images directly is analyzed.Then,a scheme of fusing the two parts based on the image geometric transformation and signal clutter ratio is proposed.The experimental results demonstrate that the proposed scheme can fusing the foreground and background images flexible and yield the final simulated RS images.