Authors: Thabit Sultan Mohammed and Nedhal Ibrahim Al-Taie
Approaches for image matching can be broadly classified into two categories: the intensity-based matching and the feature-based matching techniques. A technique from the second category is adopted in this work, where a neural framework is applied for region matching of stereo photographs. Two different types of neural networks are used in this framework, the radial basis network, (RB) for learning clustering, and the back propagation (BP) network for learning image matching.
The (RB) neural network is to cluster the regions according to the locations of their centered points. The BP network uses differential features of each region as input training data. Rather than using a single algorithm for feature extraction, the model here follows the paradigm of quality from quantity by implementing multiple feature
extraction. Features include (compactness, Euler number, and invariant moments) for each region. Results obtained from the neural networks (namely; clustering and initial matching list) are used to select the best matching pair. Lastly, a refinement step is then applied.
Keywords: Stereo image matching, BP neural network, Clustering neural network, Invariant moments and Image geometry, Neural learning, Aerial image, Computer vision.