【摘要】 为解决查询点落于分割超面以及匹配图像光照变化较大时,传统SIFT(尺度不变特征变换)算法的检测性能将会降低的问题,提出了一种改进的SIFT算法。该算法引入了灰度均匀化技术与冗余分割树。灰度均匀化技术将原图像的灰度直方图映射到更宽更均匀的直方图上,降低了光照变化的影响。冗余分割树在分割数据进行匹配时采用了2个分割超面进行数据分割,消除了查询点落于分割超面带来的影响。实验结果表明,在各种不同场景的测试下,改进的SIFT算法均提高了图像的匹配精度,性能优于传统SIFT算法。更多还原

【Abstract】 In order to overcome the disadvantages of traditional SIFT(scale invariant feature transform)which is caused by query points in the decision boundary and violent illumination changes,an improved SIFT method based on gray equalization and Spill-Tree is proposed.Gray equalization is applied into map the gray of original image to a wider and more equal gray region for reducing the impact of illumination changes.There are two decision boundaries for matching the query points in Spill-Tree which eliminates the impact caused by query points in the decision boundary.Experiment results show that the accuracy of image matching of the improved SIFT method is increased in different test scenarios and the performance is superior compared with the traditional SIFT method.更多还原