Retrieval of Petroglyph Images from a Database using Image Information

Abstract

A method of retrieving petroglyph images from a database using image information is proposed. Image preprocessing prior to retrieval is discussed, and a method for automated threshold extraction is developed. Image retrieval using a line-histogram method developed by Huet and Hancock (1998, 1998) and a centroid-radii model based on the work of Tan, Ooi, and Thiang (2000) are investigated. The centroid-radii model, when modified to accommodate missing image information and used with normalized Euclidean distance measure, was effective in retrieving similar images from a field of 437 images when the distance between images was less than 30. Preprocessing and retrieval procedures were implemented in C code. The automated threshold extraction is O(n + m2), where n is the number of pixels in the bounding box of a selected region-of-interest and m is the number of grayscale intensities in which the image is encoded; extraction of the centroid-radii distance measure is O(n), where n is the maximum dimension of a side of the bounding box, and calculation of the distance between two images in the database is O(n2) in the number of radii used to describe an image. A user survey of perceived similarity between query images and resulting retrieval sets demonstrated a direct correlation between the average distance of the retrieved image to the query image and the number of images in the retrieval set perceived as similar to the query image. A cluster analysis performed using the distance measure showed that the most meaningful clustering of images occurs at small distances.