Study of image fusion methods suitable for biomedical image

The application of image fusion in biomedical image analysis has led to a new concept about the nature of disease and to new diagnostic capabilities. However, the algorithm research of fusion is still an open research topic because the algorithm is often changed with different original images, different detectors and different research objects. This paper focuses on comparison of four image fusion algorithms based on wavelet transform to select the suitable methods for biomedical image fusion. The algorithms include: (1) weighted algorithm, (2) maximum selection algorithm, (3) stressing one image and (4) logic OR algorithm. For the sake of selecting the suitable image fusion methods for biomedical images, we propose six quantitative performance measure criterion linked with the characters of the biomedical image: standard deviation (σ), peak signal-to-noise ratio (PSNR), mean deviation (Δμ), the difference in entropy (ΔH), coefficient of correlation between the fusion image and ideal image (Correlation) and the difference in contrast between the fusion image and ideal image (ΔContrast). Using the four algorithms to process biomedical images, such as fluorescence image and the corresponding transmission image, the visual investigation and the six quantitative performance measure criterion indicate that the weighted algorithm is the most suitable method for biomedical images among the four algorithms.