论文部分内容阅读
离散小波变换(DWT)或连续小波变换(CWT)滤波后自相关运算均可对显微图像中的高频信息进行提取,依据高频能量的大小可以判断图像目标特征的离焦程度。基于上述原理,提出与小波变换相关的两类聚焦测度函数:基于 DWT 的聚焦函数、基于 CWT 滤波后自相关运算的聚焦函数。以 MEMS 器件微对准封装系统中的显微视觉单元作为实验平台,运用实验的方法确定小波基、小波因子以及小波系数的计算形式,得到可用于本显微视觉系统的两个基于小波的聚焦测度:Haar 二级小波分解系数平方和函数;尺度因子为 2-5 的 Mexican-Hat 小波滤波后自相关平方积分函数。最后利用聚焦分辨率与函数计算时间两个参数对聚焦测度函数进行量化评估。与 Brenner 函数及平方梯度函数等聚焦效果较好的基于空域聚焦测度相比:DWT 函数的聚焦分辨率为 8.43,比 Brenner 函数高 14%,其计算时间为 0.61 s,比 Brenner 函数缩短 52%;而 CWT 自相关函数在聚焦分辨率上比平方梯度函数低 41%,但计算时间比平方梯度函数缩短 36%。表明基于小波域的自动聚焦测度函数具有实用价值。
High-frequency information in microscopic images can be extracted by discrete wavelet transform (DWT) or continuous wavelet transform (CWT) filtered autocorrelation. According to the magnitude of high-frequency energy, the degree of defocus of the target image can be judged. Based on the above principle, two types of focusing measure functions related to wavelet transform are proposed: DWT-based focusing function and CWT-based filtering auto-correlation operation. Taking the micro-vision unit in MEMS micro-alignment packaging system as the experimental platform, the wavelet bases, wavelet coefficients and wavelet coefficients were determined experimentally, and the two wavelet-based focusing Measurements: Haar second-order wavelet decomposition coefficient square-sum function; Mexican-Hat wavelet filtered autocorrelation square-integral function with scale factor of 2-5. Finally, the focus measure function is quantitatively evaluated by using two parameters: the focus resolution and the function calculation time. Compared with the spatial focusing measure based on the Brenner function and the square gradient function, the focusing resolution of the DWT function is 8.43, which is 14% higher than the Brenner function. The calculation time is 0.61 s, which is 52% shorter than the Brenner function. The CWT autocorrelation function is 41% lower than the square gradient function in focus resolution, but the calculation time is 36% shorter than the square gradient function. It shows that the auto-focus measure function based on wavelet domain is of practical value.