Fuzzy correlation based algorithm for UAV image mosaic construction

Aerial image mosaic construction is very important for obtaining a wide field of view with high-resolution image. After testing several image mosaicing methods, such as descriptors based algorithms and correlation based algorithms, it was observed that the common problem of those algorithms is the occurrence of numerous erroneous associations. To address this, many strategies for identifying the correct matches were suggested, such as the Random Sample Consensus (RANSAC) method; which cannot always provide efficient results. Therefore, in our work; we have proposed to detect corners as robust features in each image; then we have developed a fuzzy matching algorithm which combines known correlation measures to provide a sufficient and precise set of correct matched features required for determining the parameters of the projective transformation model. We tried the suggested technique on many scenes and found that the results maps for well-known benchmarks and are adequate in terms of recall, precision and execution speed. The results of the developed algorithm show that it efficiently overcomes problem of false matches associated with most image mosaicing techniques.