Development of Feature Adaptation Models for Proper Registration of Aerial Images

Abstract

Unmanned aerial vehicles (UAV) have been increasingly used recently due to their high accessibility and technology growth. “Feature adaptation” is the process of finding similarities between two or more images from one scene obtained in different time intervals, viewing angles through various sensors. Application of this process in combination of obtained images and previous images can play a vital role in missions of UAV. In general, registration of two different images means finding common spots of two images considering occurred changes. It is obvious that images adaptation is one of applications of registration. All of images registration algorithms are done within 4 fundamental steps: identification and extraction of features, features adaptation, estimation of image transfer (or change) and conversion (or change in sampling rate) model. There are some constraints in each of steps that most of them are related to feature adaptation and extraction step. “Gaussian Mixture Model” has been recently employed for adaptation due to its simplicity and strength. Hence, this project has been conducted based on Gaussian model to improve quality and registration model in noisy environments toward rotation or resize