基于质量评价的无人机航拍图像拼接方法

针对大视差无人机航拍图像拼接中出现的错位、伪影问题，提出一种基于质量评价寻找最佳拼接缝的方法QEB-U。首先通过常规拼接缝估计得到初始拼接缝，然后根据无人机航拍图像的特点，综合考虑结构相似性、色差、纹理复杂度，提出一种新的质量评价函数对拼接缝上的每个像素进行评价，进而根据评价结果更新差异代价，之后再重新估计拼接缝。估计和评价过程重复执行，直到拼接缝趋于稳定则停止迭代，最后通过梯度融合生成最终结果。实验结果表明，所提方法可以避免大视差无人机航拍图像拼接中出现错位、伪影，优于目前几种无人机航拍图像拼接方法，且得到的拼接缝优先穿过道路、林地等区域，更加符合人类视觉感知，在常见的图像清晰度评价指标上表现良好。

Quality evaluation-based image stitching for UAV images

Aiming at solving problems of dislocation and artifacts in the stitching of large parallax UAV images, this paper proposed a method to find the best stitching seam based on quality evaluation QEB-U. Firstly, it estimated the initial seam by conventional seam estimation. Then, it proposed a new quality evaluation function for evaluating each pixel on the seam according to the characteristics of the UAV image like the structural similarity, color difference, and texture complexity. It updated the difference cost according to the evaluation result and acquired the seam after that. It repeated the estimation and evaluation process was until the seam became stable that the iteration was stopped. Finally, it generated the stitching UAV image by gradient fusion. The experimental results show that the proposed method can avoid dislocation and artifacts in the stitching of large parallax UAV images, which is superior to several current UAV image stitching methods. Besides, the obtained stitching seam can preferentially pass through roads, woodlands, etc. It is more in line with human visual perception and performs well in common image clarity evaluation indicators.