In this thesis, we deal with digital image sequences produced by an infrared detector array. The image sequences are characterized by large variations in noise and gray level statistics from one frame to the next. Moreover, owing to a number of defective pixels in the detector array, the frames are deliberately shifted from each other to improve the statistical properties of the signal and minimize noise. This and the inherent pointing uncertainty of the instrument result in random errors in frame positioning larger than a few pixels. This thesis discusses two point-feature based techniques to register such frames. Image registration, in general, deals with the establishment of correspondence between images of the same scene. In order to establish this correspondence, two point-feature extraction and matching algorithms have been developed and evaluated. The basic approach in both algorithms is to determine a multidimensional space of possible transformation parameters and choose a point in this space with the maximum probability of occurrence. Both algorithms make use of a localized histogram equalization and a thresholding function for feature extraction. The first relies on a priori information about the image sequences, that they are only translated from each other and not rotated, dilated or skewed, to generate a 2D space of possible shifts and a probability density function associated with this space. We then pick out the displacement with the highest probability of occurrence as our solution. The second algorithm also considers rotation by matching points in the two frames based on their relative distances from other points in their respective frames. From this initial match, we determine a 3D space of rotation and translation transformation parameters and the probability associated with each point in this space. Just as in the first algorithm, we again pick out the points that produced transformation parameters with the highest probability of occurrence.
Library of Congress Subject Headings
Infrared array detectors; Image processing--Digital techniques
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Chandrasekhar, Adith, "Point extraction and matching for registration of infrared astronomical images" (1999). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus