Analysis of Fusion Technique Using Different Wavelet Transforms

In this paper, a novel region–based image fusion technique using different wavelet transforms, that integrates multiscale image segmentation and a statistical fusion approach is considered for fusing the multisensor images. Compared to pixel-level image fusion schemes, region-based fusion schemes are less sensitive to noise. But, the region based wavelet transform technique is also vulnerable to noise, as it classifies all noise to new regions with different frequency bands. The effect of noise in the image can however be suppressed using advanced wavelet transform like dual tree complex wavelet transform and dual tree complex wavelet packet transform, which possess properties like shift invariance and directionality. In some cases, the frequency decomposition provided for the signals by the DTCWT might also be not optimal. This drawback is overcome in this paper by using the dual tree complex wavelet packet transform (DTCWPT) which provides good directionality, shift invariance and better image denoising. Performance comparison using quantitative measures like PSNR, entropy and RMSE indicates the effectiveness of the proposed method over other techniques.