A Practical Matching Confidence Analysis Technique Based on Neural Networks for Normalized Edge Magnitude Cross Correlation

Jian-xiang CHEN,Nong SANG
DOI: https://doi.org/10.3321/j.issn:1000-1328.2005.06.010
2005-01-01
Abstract:Matching confidence is important for measuring the matching quality. For normalized edge magnitude cross correlation matching algorithm, the measure of matching confidence based on neural networks was studied. The training samples used to train a BP network were the matching results with the reference image and its noisy versions. The trained network can be adopted to measure matching confidence. Experimental results with real satellite images and aerial images gave the average correct rate over 90%, and proved the effectiveness of the method.
What problem does this paper attempt to address?