Performance Assessment of Change Detection Methods Based on Neighborhood Mean Rules in Wavelet Fusion
Change detection is the process which use images of a scene captured at different interval of times. By using an appropriate algorithm, it is possible to detect the change occurred with time. This can be helpful in many areas like diagnosis of various diseases, land use change, surveillance and many other remote sensing applications. For the detection of change, output images obtained from different change detection algorithms may be fused to obtain an image which give more detailed output. This paper presents a performance analysis of various image fusion methods. The performance has been compared on the basis of accuracy and kappa coefficient. The process involves applying the two multi-temporal images to mean ratio operator as well as log ratio operator. The output obtained from the operators is fed to the DWT and SWT which convert applied images into wavelets coefficients. Then fusion rules are applied on the wavelet coefficient to fuse them together. Finally, the image is restored from coefficient map by applying Inverse wavelet transform on the coefficient map. The fused coefficients of low frequency subband are acquired by adding average value of the coefficients with the maximum value of the coefficients obtained from the wavelet transform. Similarly, the fused coefficient map for high frequency subband is obtained by taking the neighborhood mean of the coefficients. The dataset belongs to the city of Ottawa has been utilized in this paper. For comparing the performance, Kappa coefficients and Percentage correct classification has been used. The evaluation has been made based upon quantitative as well as qualitative results. The results obtained from the comparison prove the performance of SWT better than DWT.
Keywords: Neighborhood Mean Ratio, SWT, Image Fusion, Remote Sensing.