A Note on Evaluation of Image Recognition Systems
The ability to systematically evaluate the results of automated image processing systems has been problematic. It remains so. In this work we address two issues in testing: 1) the use of incomplete and inaccurate reference data; and 2) we introduce a new, somewhat faster method of evaluating systems that recognize linear structures. We illustrate the latter using road and lines-of-communication recognition systems. We describe past work on evaluation systems and compare their strengths and weaknesses relative to the current work.
KeywordsFalse Negative Rate Road Segment Versus Versus Versus Versus Versus Versus Versus Versus Versus Road Extraction
- [Forstner]Forstner, W., “10 Pros and Cons Against Performance Characterization of Vision Algorihtms”, Workshop ‘Performance Characterization of Vision Algorithms’, Cambridge, 1996Google Scholar
- [Rosenfield]Rosenfield, G.H. (1986) Analysis of thematic map classification error matrices. Photogram metric Engineering and Remote Sensing, 52,5, pp 681–686.Google Scholar
- [Wiederman]Wiederman, C., Heipke, C., Mayer, H. Empirical Evaluation of Automatically Extracted Road Axes, CVPR Workshop on “Empirical Evaluation Methods in Computer Vision”, IEEE Computer Society Press, Los Alamitos, California, 172–187Google Scholar