画像レジストレーション処理ユーザ支援システムの開発

抄録

Image registration with registers two or more images in geometric conformance is a fundamental tool of image processing. This is one of important preprocessing of image processing. For example, it is useful in change detection, stereoscopic estimation and mosaicking of several images. This paper describes the registration processing system with user support function to accomplish the high accuracy of geometry and the easy operating of the user interface. Concerning conventional registration processing system, some transformation formulas such as affine or Hermitean transformation are installed, but they are not adaptive transformations for distortion characteristics of image, So it needs the trial and error to define a transfomation formula corresponding to the distortion characteristics of the image. Then we have to select GCPs (Ground Control Point) and to input them to image processing system. In general, it is difficult to select many GCPs from satellite image, so hardly can we select sufficient number of GCPs for the registration and its estimation, and the most time-consuming part of the registration processing is the computation of the interpolation processing. Therefore we develop
image registration processing system from following two points for user interface to process registration efficiently under the interactive computer environment and the high accuracy correcting process as follows Simplification of GCP selection ; Juxtaposition of two images, input of GCP coordinates by indication devices, e.g., track ball and automatic detection of the candidate GCP. (2) Automatic estimation of distortion model ; Modeling of distortion by polynomial and defining an optimum polynomial order automatically by PSS (Prediction Sum of Square). As a result of estimation of the performance of developed system, the registration accuracy of about less than 1 pixel are obtained and it becomes about one-forth comparing as conventional system to select and input GCPs. Therefore we can develop the user supporting algorithms of less operation load for a high accuracy image registration processing.