Abstract: In the image stitching, aiming at the default that the traditional Harris and some present improved algorithms can only detect the corner point in a single scale and there may be loss of corner points, corner location offset and lack of sensitive to noise which leads to low detection rates, this paper presents an optimization corner extraction algorithm based on affinity propagation meaning, compared with bilateral filtering corner detection algorithm. The algorithm which combines bilateral filtering function with the multi-scale detection, uses the affinity propagation algorithm（AP clustering algorithm）, a new clustering algorithm, to optimize the efficiency of the candidate corner extraction. At last, this paper also presents the improved corner detection operator. The experiments are achieved in OpenCV based on Visual Studio 2010 platform. The simulation results show that the proposed algorithm can more accurately detect the image corners with better results and more practicality.

Key words: Harris algorithm, bilateral filtering, multi-scale, corner point detection, affinity propagation algorithm

摘要： 针对传统Harris角点检测算法和目前一些改进算法应用在图像拼接时，仍然可能存在只可在单一尺度上检测、角点位置不准确、伪检和对噪声敏感致使检测率不高等缺点，提出一种基于AP聚类角点提取优化的双边滤波（BF）角点检测改进算法。该算法在对图像进行双边滤波和多尺度角点检测的基础上，采用一种新型的聚类算法——近邻传播聚类算法（AP聚类算法），对候选角点提取真实角点的效率进行优化，并对角点算子进行改进。实验是在VS2010+OpenCV平台实现的。结果表明提出的改进算法不仅提高了角点提取效率，而且更加精确地检测图像角点，具有更好的效果，更强的实用性。

关键词: Harris算法, 双边滤波, 多尺度, 角点检测, 近邻传播聚类算法