Nonlinear Processing of Large Scale Satellite Images via Unsupervised Clustering and Image Segmentation

– For large scale satellite images, it is evitable that images will be affected by various uncertain factors, especially those from atmosphere. To minimize the impact of atmosphere medium dispersing, image segmentation is an essential procedure. As one of the most critical means of image processing and data analysis approach, segmentation is to classify an image into parts that have a strong correlation with objects in order to reflect the actual information collected from the real world. The image segmentation by clustering basically refers to grouping similar data points into different clusters. In this article, an unsupervised clustering technology is proposed for processing large scale satellite images taken from remote celestial sites where none explicit teacher is introduced. As an effective approach, K-means clustering method requires that certain number of clusters for partitioning be specified and its distance metric be defined to quantify relative orientation of objects. Then image processing system forms clusters from input patterns. Diversified large scale image features are investigated using unsupervised methods. At the same time, to limit computational complexity for real time processing consideration, a simple study is also conducted where tristimulus values are selected to represent three-layer color space. Simulation results show that this approach is very successful for spatial image processing. Key-Words Large scale satellite image, K-means clustering, Image segmentation, Nonlinear processing.