相机移动场景下的多曝光图像融合系统设计

Design of a multi-exposure image fusion system under the moving camera scene

由于相机和显示设备的动态范围远小于人眼可识别的动态范围,相机获取到的图像往往不能兼顾亮部细节与暗部细节。多曝光图像融合可以解决上述问题,但目前该方法大多基于相机与目标场景相对静止的前提。文中通过结合图像配准和图像融合,实现一个可以处理运动相机拍摄的多曝光图像的融合系统。首先,配准模块提取SURF特征点,进行欧式距离粗匹配;其次,配准模块利用RANSAC算法进行精匹配,提取变换模型参数并进行投影变换以矫正图像;最后,融合模块利用金字塔融合法融合已配准图像。实验结果表明该系统可有效矫正图像空间位置偏差,扩大图像的动态范围,提高图像质量。

The dynamic range of cameras or display devices is often lower than that of human eyes, therefore images captured by normal cameras are often lack of details in the dark or highlight areas, which can be addressed by multi-exposure image fusion methods. However, almost all these methods are based on the premise that the camera should be relatively static to the target scene. In this research, a system that can proceed images captured by a moving camera is proposed. Firstly, the registration module extracts SURF features from the images and performs a rough match based on Euclidean distance.Secondly, the registration module matches the feature points precisely by using RANSAC algorithm, computes projective transformation model and implements registration. Finally, the fusion module fuses the registered multi-exposure images by using multi-resolution pyramid fusion method. Experimental results indicate that the proposed system can rectify the spatial deviation, expand the dynamic range of fused image and improve its quality effectively.