Please use this identifier to cite or link to this item: https://hdl.handle.net/11681/11337
Title: Development of computer vision techniques for automatic feature extraction
Authors: Gordon, Daniel K.
Pascucci, Richard F.
Keywords: Computer vision
SAR Imagery
Descriptor sets
Automatic feature extraction
Expert system
Publisher: U.S. Army Engineer Topographic Laboratories.
Engineer Research and Development Center (U.S.)
Series/Report no.: ETL ; 0451.
Description: Technical report
Abstract: In previous work, 52 descriptors (feature identifiers) and 501 descriptor sets were identified as being used by image analysts for the characterization of features found in radar imagery. In the research investigation described herein, the descriptor sets were tested and validated. Following this, computer vision techniques were identified and developed to automatically recognize these descriptor sets. The identification procedure includes image preprocessing (e.g., edge enhancement, density slicing, neighborhood encoding and thinning), raster to vector conversion, and the processing of the resultant vector data (e.g., identification of points, lines, and areas, referred to as primitives, including a description of primitive size, shape, position and orientation). Finally, the relative positions of primitives were examined, and the similarity between groups of primitives and descriptor sets was quantified. The images used were softcopy versions of graphic, line-drawn examples of the descriptors that were identified in the previous work. The computer vision techniques that were developed have been demonstrated successfully in a tightly controlled environment, on images containing little extraneous information. Research is currently being expanded to include carefully selected, uncluttered radar examples. The final goal of the investigation is the automatic identification of selected features from images acquired under a variety of conditions.
URI: http://hdl.handle.net/11681/11337
Appears in Collections:Documents

Files in This Item:
File Description SizeFormat 
ETL-0451.pdf1.79 MBAdobe PDFThumbnail
View/Open