Novel Fusion Rules for Discrete Wavelet Transform based Image Fusion

Objective: The accuracy and reduction in speckle noise is an issue of major concern in change detection methods. In this paper, new fusion rules for Discrete Wavelet Transform (DWT) based image change detection have been proposed. Method: Multi-temporal images have been applied to Log Ratio and Mean Ratio operators to generate the source images. Both the source images are decomposed into wavelet coefficients through DWT. The fused image is obtained by applying the proposed fusion rules on the decomposed wavelet coefficients. The fusion rules for low frequency sub-band is based on addition of the average and maximum value of the wavelet coefficients while the fusion rule for high frequency sub-band is based on the neighborhood mean differencing of the coefficients. Findings: The difference image is generated by applying inverse wavelet transform on the fused coefficient map. The changed and unchanged areas have been classified by Fuzzy C Means (FCM) clustering. The results have been compared based upon parameters like Overall Error (OE), Percentage Correct Classification (PCC) and Kappa Coefficient (KC). The qualitative and quantitative results show that the proposed method offers least overall error. The accuracy and Kappa value of proposed method are also better than its preexistences. Application: The method has applications in remote sensing, medical diagnosis and disaster management.