Image Enhancement via Multiple Canonical Correlation Analysis

For image understanding and performing detection, recognition and identification functions, different features representing the scene should be extracted from different images of the scene. From the images of the scene captured by different sensors, such as operating at different bands, different features can be obtained. For sensor fusion, first the difference in the information content of these separate data should be assessed. In this study, demonstration of the use of Multiple Canonical Correlation Analysis (MCCA) for information extraction from the multi-sensor data is provided. From the registered data captured with three different cameras, multiple images are obtained by pixel shifting methodology and analyzed via MCCA. The scene details are obtained from the canonical variates and the level of mutual information of these new data sets is determined via canonical correlations.