基于DCT变换的图像融合方法研究

Image fusion algorithms using discrete cosine transform

提出了一种基于离散余弦变换（DCT）以及一种结合小波变换与DCT变换的图像融合新方法。前者将源图像进行分块DCT变换，依据DCT系数的高频能量，对源图像的对应区域进行融合。后者利用DCT系数的高频能量对小波分解后得到的低频予幽进行融合，同时以此为依据对小波最高分解层的小波高频系数进行选择，其他分解层的小波高频系数依据最大局部方差准则进行融合。依照平均误差、峰值信噪比以及均方根误差等客观评价标准，将新方法与其他常用的基于小波变换或DCT变换的融合方法进行了比较。实验结果表明，结合小波变换与DCT变换的图像融合新方法获得的融合效果优于其他方法。该疗法与常用的基于小波变换的融合方法相比，其平均误差减少了40．8％～69．5％，峰值信噪比提高了9．9％～15．6％，均方根误差减少了34．8％～47．5％，评价结果与日视效果相吻合，表明该方法能有效地提高图像融合的质量。基于DCT变换的图像融合新方法的融合效果仅次于结合小波变换与DCT变换的图像融合新方法且其计算量相对较少，适用于实时处理。

An image fusion algorithm based on discrete cosine transform （DCT） and an image fusion scheme using wavelet transform combined with DCT were proposed. The former based on DCT fused the corresponding areas of the original images according to the DCT coefficient high frequency energy and the latter using wavelet transform and DCT fused the low frequency submaps by use of DCT coefficient high frequency energy. The wavelet high frequency coefficients of the highest wavelet decomposition level were decided by the wavelet low frequency coefficients. The wavelet high frequency coefficients of the other wavelet decomposition levels were selected with the greater local deviation. The fused images of the algorithms purposed were evaluated by some parameters such as average error, peak signal noise rate, and root mean square error, compared with other conventional image fusion methods based on wavelet transform or DCT. The experimental results show that the new algorithm based on wavelet transform combined with DCT provides the best performance with 40.8 % to 69.5 % reduction in average error, 9. 9% to 15. 6% improvement of peak signal noise rate and 34. 8G to 47.5% reduction in root mean square error. It is superior to the conventional methods using wavelet transform. This algorithm improves the quality of the fused image effectively, the evaluation results coincide with the visual effect very well. The algorithm based on DCT needs less computational burden and is more suitable for the real-time processing.