Knowledge-Based Scale Transfer Approach for Image Fusion

The purpose of image fusion is to integrate images with different resolutions or even images from different sensors, so that the identifiability and reliability of images are increased. In order to improve image identifiability, a knowledge-based scale transfer (KBST) fusion technique was developed. Knowledge-based and scale transfer were the major concepts of this proposed approach. Firstly, SPOT multispectral (XS) images were used for major categorical (water, vegetation, and bare soil/built-up) landcover classification of the studied area for acquiring spectral knowledge of landcover. Regression relationships between digital panchromatic (PAN) images and XS images were then established and were used for subsequent scale transfer. The class-specific regression models help preserve spectral information during scale transfer. Finally, a scale transfer algorithm using class-specific regression models was adopted for fusing a SPOT XS image and a PAN image, yielding a multispectral and high-resolution image which offers more details of the studied area than other spatial domain fusion techniques (the linear combination method and the square root of Pan x XS; method). Therefore the images fused by our KBST fusion are superior to those fused by other methods studied in this work in terms of visual and statistical assessments.