Geometric Correction of Digitized Art

Abstract

The development of digital image databases of art requires large numbers of photographed art images to be scanned into electronic form. These scanned images often contain perspective distortions caused by a misalignment of the art object and image planes which can be corrected using a planar perspective transformation provided that correspondence between four points in the distorted image and a known (undistorted) rectangular region can be established. A method of locating the four corner points of art objects is presented, based upon a global analysis of a low resolution version of the image followed by a high resolution analysis of four local regions of interest. A feature hierarchy consisting of line equations, corner locations, and convex quadrilaterals is formed on the low resolution image. The set of quadrilaterals is analyzed to find the outline of the art object within the image, and the four vertex locations of the chosen quadrilateral are then used to extract four high resolution regions from the distorted art image, whereupon the precise location of each of the four corners is determined. The technique is shown to work on a variety of images of Northwest Coast Native Art; it is effective for locating the corners in very large format digital images due to the use of global analysis at low resolution followed by local analysis at high resolution.