A study on image processing with multivariate statistical dimension reduction approach

Analysis of multispectral image gained significant importance with the rise in the amount of studies in the areas of mining engineering, geodesy, photogrammetry engineering and esspecially geographical information systems. The combined usage of multivariate statistical analysis and image processing is unavoidable since different data structure is obtained from each one of the multispectral image. Principal Components Analysis which is a multivariate dimension reduction methodology produces the principle component images on which observations and interpretations can be easily performed, by succesfully reducing the number of layers that are worked on. By classical principles components analysis the obtained eigenvalue and eigenvector structure can be interpreted by using the graphics of the loads and scores in the analysis of the multispectral images. In multi spectral satellite images, classification is expressed as the process of determining which feature group each pixel value in the related image belongs to. This classification process is also expressed as classifying images by identifying similar geographical features in the related image. In the classification process, it is mostly done by considering the spectral reflections of the objects. The main purpose of classification is to group images with similar spectral characteristics. Within the scope of this study, the analysis and comparison of LANDSAT-7 satellite images of the province of Izmir were made by using the main component analysis, which is a multivariate statistical dimension reduction method. In addition, Principal Component Analysis was applied to examine the geographical structure in the selected regions. In the last stage of the application, a new approach has been put forward to classify images that are similar to each other by size reduction in the analysis of satellite images by applying Factor Analysis to the same images. Keywords: Multivariate Image Processing, Image Enhancement, Principal Component Analysis, Factor Analysis.