Investigates Image data fusion techniques that combine image and track data from multiple sensors to achieve improved accuracies and more specific inferences than could be achieved by the use of a single sensor alone. Our aim is to explore the state-of-the-art image processing algorithms for achieving effective data fusion as in:
CCD-Thermal IR Image Fusion
1. Introduction
- This algorithm is a multi-sensor data fusion method for visible imaging source (CCD) and thermal imaging source (LWIR:Long Wavelength Infrared). CCD sensor has advantages for texture information but has limitations for detecting objects with low illumination environment. On the contrary, thermal IR sensor does not experience the low illumination problem, but texture information cannot be acquired. Our research subject is that finds complementary image by fusing each image data from CCD and thermal IR sensors.
2. Main algorithm and principle
- General CCD-Thermal IR image fusion algorithm structure
1. Register two input images to have same field of views.
2. Generate fused image by multi-scale transform (MST) or Non-multi scale transform method.
3. Apply color reconstruction processing for natural image.
- Proposed CCD-Thermal IR image fusion algorithm
1. Image registration by using adaptive Harris corner detection and RANSAC algorithm
2. Novel feature matching based on Normalized Mutual Information (NMI) and edge orientation coincidence (EOC)
3. Conduct thermal IR image transformation before pixel data fusion
4. Pixel intensity fusion by weighted averaging method
3. Experiment Results
4. Demo
