基于双正交小波变换与特征匹配的脊柱序列自动拼接

Spinal Sequence Automatic Stitching Based on Biorthogonal Wavelet Transform and Feature Matching

目的:本文提出了一种脊柱X线图像序列全自动无缝拼接方法。方法:首先采用双正交小波变换对图像序列进行多分辨率分解,并结合Canny算子提取图像的有效边缘轮廓,通过计算边缘轮廓矩阵E与和值矩阵H获得图像特征点;其次利用归一化互相关(normalized cross correlation,NCC)算法实现特征点粗匹配,并引入稳健的随机抽样一致(random sample consensus,RANSAC)算法去除误匹配对,实现图像精匹配;然后利用改进的遗传算法对图像序列自动排序,实现匹配系统的自动拼接;最后利用加权平均融合算法实现图像的平滑无缝拼接。结果:实验结果表明该算法可实现高质量快速的脊柱X线图像序列自动拼接。结论:该算法对X线弱对比度脊柱图像序列具有很强的鲁棒性,在一定程度上克服了X线图像序列噪声强、灰度集中、边界模糊、重叠面积过大的缺点及位置关系的限制,具有较强的应用价值。

Objective： An automatic seamless stitching method with spinal X-ray image sequence is presented in this paper. Methods： First, biorthogonal wavelet transform is used to implement decomposing of the multi-resolution and the effective edge of the image can be extracted by this method combined with Canny operator. The feature points of the image can be obtained by calculating the edge contour matrix E and the value matrix H. Second, the roughly matching of feature points can be achieved by using Normalized Cross Correlation （NCC） algorithm and the random sample consensus （RANSAC） algorithm is introduced to remove false matching pairs and to achieve precisely matching. Third, the image sequence is automatically sorted with the improved genetic algorithm to achieve automatic stitching. At last, the weighted average fusion algorithm is applied to achieve smooth and seamless image stitching. This algorithm is robust for the weak-contrast X-ray image sequence. Results： Experimental results show that high-quality and fast image sequence stitching can be obtained automatically by using this method. Conclusions： To a certain extent, it overcomes the shortcomings of X-ray image sequence such as the strong image noise, concentration of values of pixels, blurred boundaries, large overlap area and the sequence constraint, and therefore it may be applied to in medical imaging field widely.