针对图像拼接技术中特征精匹配耗时长,图像配准精度低导致拼接区域模糊等问题,提出一种基于改进网格运动数据和加权投影变换的图像配准算法。该方法使用ORB(Oriented FAST and Rotated BRIEF)算法进行图像特征提取,再利用暴力匹配算法进行图像粗匹配。然后将图像划分成多个方形网格,进行网格特征数量统计,通过计算五宫格特征分数来剔除错误匹配,得到精匹配特征点集。最后通过引入距离权重系数获得加权投影变换模型实现图像配准。将本文算法与其他方法在拼接序列集上进行测试比较,实验结果表明,本文算法在配准精度上平均提高28.7%,在特征精匹配速度上提升43.6%,拼接的全景图像无明显几何错位和畸变,整体成像自然。

Fine feature matching in image stitching and blurred panorama areas is time consuming due to inaccurate image registration. To mitigate this issue, this study proposes a new image registration model based on improved grid motion statistics and weighted projection transformation. The method uses the oriented fast and rotated BRIEF (ORB) algorithm to extract and describe the image features. The brute-force matching algorithm is used for rough image matching. The image is divided into multiple square grids. Then, these grid features are counted and five grids feature scores are calculated to eliminate error matching and obtain the refined matching feature set. Lastly, image registration is achieved by adding a distance weighting coefficient to construct the weighted projection transformation model. Comparing the proposed algorithm with other methods used in the stitching sequence set, the experimental results revealed that the accuracy of the proposed algorithm was improved by an average of 28.7% for image registration and the feature matching speed was improved by 43.6%. The stitched panoramic image did not show any obvious geometrical dislocation or distortion, and the overall imaging appears quite natural.