|
The task of automated image registration can be divided into two major components: 1) the extraction of features from images; and 2) the search among the extracted features for the matching pairs that represent the same object in the field of view of the images to be matched. A new method was introduced for automated registration of DEM (Digital Elevation Model) images with different noise levels and amounts of missing data. With this new method, instead of using isolated points of interest, edges in a generalized sense are first extracted from the DEM images as the features on which the whole registration process is based. The edges are extracted with multiple variance filters of different widths followed by a linear feature extraction method based on local maximum convolution. As the second step, the shape of the extracted edges is represented with a rotation-invariant code that is generated dynamically for each sub-section of an edge. Then the similarity of two edges, one from each image, is evaluated with a parameter that reflects both the length of the edges and the linear correlation coefficient calculated from their rotation-invariant codes. When the matching edge pairs are determined, the corresponding pixels within each edge pair are used as "tie points" to construct a redundant equation set for a least squares solution for the projection parameters to merge one DEM image into another, or one image of any type that is registered with the first DEM into another that is registered with the second DEM. For this pair of images, a total of 856 pixels from four matched edge pairs served as "tie points" to form 1712 equations, from which four projection parameters (one scaling, one rotation and two translations) were obtained. The entire process requires little human intervention. As a quality check, after projecting one DEM into the other, the difference between the two corresponding pixels was calculated, if there is no missing data in either image. Another example of a registered DEM pair and the corresponding mosaic image of a radar image and a digitized aerial photo are shown on separate pages. Main Page || Original DEM Images || Extracted Features || DEM Mosaic I
|
|
|
