Effect of noise on wavelet transform based image fusion algorithms

Image Fusion is used to integrate multiple images into a composite image, which contains complementary information from each of the source images. In defense applications, fusion is widely employed to obtain images pertaining to the object under surveillance and also for mapping terrain for navigation purposes. Surveillance imaging generally use two imaging sources, one an Infra Red (IR) camera and the other a conventional digital camera; and the images are usually captured under low-lighting and night time conditions. The imaging process is prone to sensor noise which degrades the fusion performance. This is because the noise is also considered as useful information by the fusion process, resulting in a corrupted fused output, rendering the image useless. We have formulated different fusion techniques based on the Discrete Wavelet Transform (DWT) and the Principal Component Analysis (PCA). In this paper, the efficiency of these fusion algorithms is evaluated under the presence of sensor noise. The Structural Similarity Index (SSIM), the Peak Signal to Noise Ratio (PSNR) and the Root Mean Square Error (RMSE) are used as metrics to evaluate the performance of the fusion schemes. Our experiments have shown that the DWT based fusion method that utilizes the energy of the wavelet coefficients for fusion produces good results under the noise constraints imposed.