快速SUSAN-Laplace角点算法  

A fast Laplace-SUSAN comer detector

作　　者：吕冀[1] 汪渤[1] 高洪民[1] 

机构地区：[1]北京理工大学信息科学技术学院,北京100081

出　　处：《光学技术》2008年第6期873-876,共4页Optical Technique

基　　金：国家部委预研资助项目(51405030104BQ0171)

摘　　要：提出了一种新的SUSAN-Laplace角点算子。新算子首先使用拉普拉斯-高斯核对图像进行卷积,并将图像转换为二值或三值图像,然后使用SUSAN算子检测角点。这样既快速又具有SUSAN算子的位置精确性。SUSAN- Laplace角子具有较好的旋转、仿射、光照、尺度和噪声不变性,其计算速度也较快,比Harris角点快1.5倍,比SUSAN角点快6倍。应用中SUSAN-Laplace角子有一种快速算法,快速算法在计算拉普拉斯-高斯卷积时使用移位运算代替全部乘除法运算,能进一步提高计算速度。SUSAN-Laplace角子较其它角点算法能产生更丰富的角点,应用到SIFT算法中可提高对小目标的识别能力。A novel Laplace-SUSAN corner detector is described. The original image Convolved by a Laplacian of Gaussian filter is translated to a binary or ternary image according to threshold, in which the algorithm uses the SUSAN detector to detect comers. The Laplace-SUSAN detector has excellent high-speed and location accuracy, that can perform 1.5 times faster than Harris and 6 times faster than SUSAN and whose corner is shown to be invariant to image rotation and scale and robust across a substantial range of affine distortion, addition of noise, and change in illumination. Its fast algorithm also is provided, that can substitute all multiplicative operations with shift operations for the Laplacian of Gaussian smoothing. The Laplace-SUSAN detector has a lot more corners than other detectors, so applied to SIFT it can perform more excellent for small object recognition.

关 键 词：图像处理 目标分类 角点检测 拉普拉斯-高斯核 SUSAN 二值或三值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]