用于目标识别的PCA-SC形状匹配算法

PCA-SC shape matching for object recognition

基于形状上下文(Shape Context)算法并融合主成分分析(PCA)的降维思想,提出了一种PCA-SC算法来提高形状匹配和目标识别的速度和抗噪能力。该算法将SC算法获取的特征矩阵构成协方差矩阵,按照特征值由大到小的准则进行降维,形成新的特征矩阵用于匹配和识别,既抑制了噪声干扰,提高了识别准确率,又能够提高匹配速度,易于满足工程应用对实时性的要求。利用MNIST图像数据库中的图像进行了实验分析,结果表明,PCA-SC算法在保持了SC算法原有的定位准确、抑制噪声等优点的基础上,识别速度提高了1倍;准确率达到了96.15%,提高了约0.5%;而且抗噪性更强,可用于匹配和识别较复杂的形状和目标。该算法基本满足匹配和识别对速度、准确率和抗干扰性等方面的要求。

A new algorithm based on Shape Context（SC） and Principal Component Analysis（PCA） called PCA-SC was proposed to improve the matching efficiency and anti-noise performance in shape matching and object recognition. The algorithm establishes a covarianee matrix based on the feature matrix obtained by the SC, then reduces its dimensions according to the size of ei/~en value and forms a new feature matrix to implement the shape matching and object recognition. The proposed algorithm can not only remove noise interference and improve the recognition accuracy, but also can enhance the matching efficiency for real-time application. The experimental results of MNIST database indicate that the PCA-SC algorithm outperforms previous SC algorithm, and its recognition speed is doubled that of SC and the accuracy reaches to 96.15% increased by 0. 5%. Furthermore, the anti-noise per- formance becomes stronger. Therefore, this algorithm shows better performance for shape matching and object recognition in efficiency, accuracy and anti-noise.