Metric Aspects of Digital Images and Digital Image Processing.
Abstract
The ability of human pointing in digitized aerial imagery is studied. Automated methods for extracting edge locations in digital imagery are determined with subpixel accuracy. An algorithm for automatically extracting cross targets from aerial imagery is developed and implemented. Crosses are recognized using a method based on Fourier descriptors. Precise locations of the crosses are determined by Fourier descriptions and further refined using a least squares adjustment model. Mean cross location errors are as low as 0.03 pixels. Metric distortions caused by digital image processing techniques are studied using the Fourier descriptor-least squares algorithms. Cross positions are 'shifted' by as much as 0.5 pixels when cosine compression is applied with a 16-fold bit reduction. Mean and median filters show little metric distortions. An isotropic median filter is developed. Resampling with various interpolation functions shows that metric distortion is lower when using bilinear interpolation as compared to nearest neighbor or bucubic interpolation.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1984
- Accession Number
- ADA147211
Entities
People
- E. M. Mikhail
- O. R. Mitchell
Organizations
- Purdue University