Image Fusion Based on Markov Random Field Model
JIANG Zhi-yang;College of Electronics and Information in Tongji University;
Traditional fusion rules in image fusion are max rules and weighted average rule pixel by pixel. The former is able to preserve the textures while irresistance to noises; the latter is advantage to process the image with noises, but loses the high frequency information. Recently rules based on region have been improved. The fusion take the neighboring pixels account and fuse image segment by segment. However the correct segmentation and the inconsistent fusion weights between regions become problems. This paper designed a model based on MRF, making optimal estimation of pixels weight, with the saliency of the pixels and the connection among pixels as constraint. The experiment performs well.
【CateGory Index】： TP391.41