The traditional LBP feature for target recognition mainly depends on local LBP histogram,but it is only fit for the calculation of LBP histogram features under the small neighborhoods and small sample points. When calculating the contrast of the pixels with more sampling points in larger scale,the dimension of the histogram will increase exponentially,which will result in the curse of dimensionality. In order to solve this problem,spatial pyramid pooling method is used for pooling LBP features,and LBP features are calculated by several neighborhood scales and different sampling point numbers,so as to establish a complete image feature descriptor. To train recognition template based on support vector machine (SVM) or other training network,the input feature sets need to have the same dimension,and the output characteristic dimension has fixed length,which can effectively avoid image distortion and information loss. The experiments prove that the proposed method can avoid the curse of dimensionality and improve the detection rate and recognition accuracy more efficiently.
参考文献
相似文献
引证文献
引用本文
郭少军,陆斌,娄树理.应用空间金字塔池化LBP特征的舰船检测识别[J].激光与红外,2017,47(6):783~788 GUO Shao-jun, LU Bin, LOU Shu-li. Ship detection and recognition based on spatial pyramid pooling LBP feature[J]. LASER & INFRARED,2017,47(6):783~788