CAJ | 학술논문

为了改进噪声鲁棒性和定位准确性，利用各向异性高斯方向导数滤波器，提出多方向角点检测算法．该算法利用一组各向异性高斯方向导数滤波器对输入图像进行卷积处理得到各个方向的滤波器响应．对于每个像素点，利用它与周围邻近像素点的滤波器响应的相关信息构造局部自相关矩阵，然后根据自相关矩阵归一化特征值及像素点处各方向滤波器响应，作阈值处理和非极大值抑制处理判定像素点是否为角点．实验结果表明，在无噪声和噪声的条件下，提出的检测方法与各向同性高斯核函数的Harris算法相比，配准角点数均提高6.0％左右，具有更好的检测性能．
위료개진조성로봉성화정위준학성，이용각향이성고사방향도수려파기，제출다방향각점검측산법．해산법이용일조각향이성고사방향도수려파기대수입도상진행권적처리득도각개방향적려파기향응．대우매개상소점，이용타여주위린근상소점적려파기향응적상관신식구조국부자상관구진，연후근거자상관구진귀일화특정치급상소점처각방향려파기향응，작역치처리화비겁대치억제처리판정상소점시부위각점．실험결과표명，재무조성화조성적조건하，제출적검측방법여각향동성고사핵함수적Harris산법상비，배준각점수균제고6.0％좌우，구유경호적검측성능．
A multi-directional corner detector using the anisotropic Gaussian directional derivative filters is proposed to improve the noise-robustness and the localization accuracy .T he proposed algorithm first uses a set of anisotropic Gaussian directional derivative filters to smooth the input image .For each pixel , its filter responses and those of its neighborhood are used to construct the auto-correlation matrix ,then a corner measure combing the normalized eigenvalues and the responses at the pixel is calculated and threshold and non-maximum suppression are used to decide corners .The experiments show that ,com-pared with Harris algorithms ,the correct detected corner numbers by the proposed algorithm increase by about 6.0% for both noiseless and noise images .