Computational Vision Framework for In-Line Prediction of CMOS/CCD Imaging Performance via Spatial Frequency Response and Its Application to Nano-Optoelectronics

The relentless miniaturization of consumer electronics drives the demand for high-performance micro-camera modules, where imaging quality is critically dependent on precise lens-sensor alignment. Conventional methods for assessing optical performance, such as spatial frequency response (SFR) measurements, are accurate but inherently offline and slow, creating a bottleneck for mass production. This work introduces a rapid, in-line inspection methodology that leverages computational vision to predict imaging quality through the quantification of lens centration error. We establish a statistical baseline using traditional SFR evaluation and then employ a telecentric imaging system to extract geometric features and measure assembly misalignments. A linear regression model reveals a significant negative correlation between the measured centration error and the SFR value. To validate industrial applicability, we conducted an online inspection of 300 consecutively manufactured modules. The results showed that modules with good concentricity, defined as having an alignment error of 25 micrometers or less, exhibited SFR values that were on average 15% higher than those of poorly aligned modules. The ability to correlate structural precision with functional optical output provides a scalable and efficient pathway for quality assurance in the production of next-generation devices, including integrated photonic circuits and metasurfaces, bridging a critical gap between macro-scale assembly and nanoscale precision requirements.