Classification Packages for Satellite Image Processing

Abstract

Automatic image segmentation and further analysis of the image is the growing need in satellite image processing eld. Segmentation involves partitioning an image into a set of homogeneous and meaningful regions, such that the pixels in each partitioned region possess an identical set of properties or attributes. Further the set of regions of interest in the image undergoes subsequent processing such as edge detection, vector conversion, area description etc. Enhanced seeded region growing is the proposed approach for image classi cation in a su- pervised method. It makes use of ICT to generate the training site. The latitude-longitude information is captured and passed through GPS enable device. These points are marked as seed points on a geo-referenced image. The algorithm uses 8-connected neighboring concept for classi cation. Once the classi cation is done the area of each class is identi ed. For further geo-spatial analysis of the image it is converted into the vector format such as KML and CSV through edge detection. The image analysis part includes the implementation of various vegeta- tion indices to nd out the amount of vegetation in an image, saving spectral and spatial subsets from an image, Contrast enhancement techniques, Conversion of ASCII les into Bitmap, and layer stacking. The efficiency is measured in terms of the kappa co-efficient, which is 0.81 for the proposed method. It requires less ground truth information compared to Maximum Likelihood (MXL) and Arti ficial Neural Network (ANN). The efficiency is better than parallel piped classi fication technique and comparable with MXL and ANN. It is faster than ANN.