| Popis: |
Satellite Image Processing has become very important in many applications. Each satellite approximately collects images of over 800,000 km2 of area per day. These images have great chances of getting blurred and distorted. To get maximum reliability of information over an landscape, the satellite collects repeated data and images of the surface using various modes of image acquisition and different types of sensors. These images might be in different orientations, scaling, illumination and affine transformation. This eventually leads to increase of redundancy among the data represented by each of the images of the same place. Therefore, a need for a technique that reduces the repeated and redundant data and converts the multiple snapshots into a single meaningful image is of critical importance. This also reduces the size of the data that is downloaded. Having these considerations, an Image registration technique with special reference to satellite images and is immune to changes in feature discrepancies is proposed. In this dissertation, the proposed image registration technique is simulated and validated using non-satellite images for better visualization of image features which can be similarly applied to satellite images also. SIFT algorithm is used to detect the distinctive features in an image, that is invariant to scaling, translation, rotation and to some level invariant to illumination and affine transformation. Thus, the features obtained by this algorithm are used for reliable matching. Prior to the image registration, detection and matching of the image features is required. Dot product of descriptors is used to perform the feature matching in the images. The retention of the match is evaluated based on the cosine angle of the dot product. Clustering Algorithm is adopted as the registration technique in the proposed algorithm and is implemented using MATLAB. Master of Science (Computer Control and Automation) |
