Reconstruction of Temporal Images by Gradient based Sequential Prediction

The identification of the factors involved in change detection could lead to a comprehensive understanding of real changes and non-real changes on a broad scale, as well as prediction capability. As a huge amount of remotely sensed data is available, most of the applications require the interpretation of images collected over a period. Frequently collected satellite images mostly present strong spatial redundancies for real changes such as deforestation, urbanization, flood, bushfire etc. and non-real changes due to various system factors or environmental noise such as illumination variation and atmospheric effects. In this case, the pixel values of two images are not same. Therefore nonlinear regression prediction model such as gradient adjusted temporal prediction procedure is applied to predict a temporal image for detecting the types of changes have occurred and is presented in this paper. As the changes are detected iteratively, the whole process converges towards the final model that better defines the temporal correlation between two adjacent images.