摘要
针对常规综合鉴别函数制作的复合匹配滤波器,对不同旋转角度的训练样本输出的相关峰值波动性较大,不便于后续探测和处理等问题,对常规综合鉴别函数到编码匹配滤波器的编码转化过程进行分析,找到导致相关峰值波动的原因,提出改进方法,即在计算综合鉴别函数的线性组合系数时将编码运算也考虑进去,并根据滤波器对各训练样本相关峰值的倒数来确定线性组合系数。计算机仿真结果表明:改进后的方法降低了复合滤波器对不同旋转角度样本输出相关峰值的波动性,提高了相关峰自动判别的准确性。
To overcome the great correlation peak fluctuation caused by the composite matched-filter prepared with conventional synthetic discriminant functions, the conversion process of coding from conventional synthetic discriminant functions to the coding matched filter is analyzed. The reason for correlation peak fluctuation was found. A modified scheme, in which the coding arithmetic is considered when the linear combination coefficients of the synthetic discriminant functions are calculated, is proposed. The results of the computer simulation indicate that, compared with the conventional synthetic discriminant functions, this improved method can greatly reduce the correlation peak fluctuation of the composite matched-filter, simplify the post-processing and improve the veracity of the automatic identification of the correlation peak.
出处
《应用光学》
CAS
CSCD
2007年第3期376-379,共4页
Journal of Applied Optics
关键词
旋转不变模式识别
常规综合鉴别函数
训练样本
rotation invariant pattern recognition
conventional synthetic discriminant function
training sample
