影像融合在近年來對於電腦視覺的領域當中,是一個非常重要的議題,其應用面非常廣泛,影像融合技術能夠將多張相同視景的影像融合成單一更富含資訊且清晰的影像;在醫學領域方面CT影像主要能掌握結構性的組織,如:骨骼,而MRI影像則較能掌握功能性的組織,因此若能將CT與MRI做很精確且有效的融合,融合後的影像將能輔助醫生做更精確的診斷,且降低誤判的機率。在這篇論文中,我們提出來一個新的影像融合的方法,在頻率域裡面分別針對高頻部分及低頻部分採用不同的方法進行融合,再將高頻部分及低頻部分得到的係數進行小波逆轉換進而得到融合後的影像。其中低頻的部分我們採用梯度金字塔,而高頻的部分則採用鄰域方差。本研究與現存拉普拉斯融合法進行實驗比較,並且使用熵值、峰值信號雜訊比、相關係數三種不同的評估指標來進行分析,實驗證明我們的方法保留了更多的訊息並降低雜訊,且更完整保留原醫學影像中的重要資訊,證明我們的方法確實對CT與MRI較其他方法有更佳的融合結果。
Image fusion is a very important issue in the field of computer vision in recent years, and its application is very extensive. Image fusion technology can fuse multiple images with the same scene into a single clear image with more information. In the field of medicine, CT images mainly can master the structure of the tissue, such as a bone, while MRI images are more able to grasp functional tissue. So if CT and MRI can be fused very accurately and effectively, then the fused image will help doctors the more accurate diagnosis, and reduce the probability of misjudgment. In this paper, we proposed a new method of image fusion. In the frequency domain, we use different methods for the high frequency part and the low frequency part, respectively. The coefficients obtained from the high frequency part and the low frequency part are converted by inverse wavelet transformation to obtain the fused image. We use gradient pyramid for the low frequency part and neighborhood variance for the high frequency part. This study is compared with the existing Laplacian fusion method, and uses entropy value, peak signal noise ratio and correlation coefficient to analyze the performances. The experimental results showed that our method retains more information and reduces noise. Beside, our method provides more complete preservation of important information in the original medical images. These experimental results proved that our method outperform other method in the fusion of CT and MRI images.