When modulation transfer function (MTF) detection for Bayer filter color spatial cameras with low modulation transfer function is conducted, the spatial frequency cutoff will occur, and the complete MTF curve at zero to Nyquist frequency cannot be obtained. Meanwhile, edge detection and edge spread function processing are the two most important steps in edging methods. The ISO 12233 method employs the centroid method for edge detection, which is greatly affected by noise. Canny operators have low sensitivity to noise and are widely adopted in image edge detection. However, they have some limitations in edge detection in MTF tests. Additionally, the ISO 12233 method requires the sampling rate of the edge spread function to be as integer as possible. Otherwise, the edge spread function curve will not be smooth and continuous, and will not accurately represent the edge spread function distribution of the actual image. Moreover, since the subsequent differential and FFT processing will amplify ESF data fluctuations, it is necessary to homogenize ESF data. However, the present homogenization method will introduce new errors which cannot be ignored, resulting in insufficient accuracy of the algorithm.

To meet the requirements of Bayer filter color cameras for high-precision MTF measurement, we improve the calculation algorithm and detection method. The MTF detection of three primary color components is proposed, and the color image test results are obtained by weighting according to the calibration results of the light source. Additionally, the edge method is optimized by improving the two key steps of edge detection and edge extension function data processing. Firstly, the image edge fitting is carried out by employing the Canny operator with strong noise resistance. When the Canny operator is utilized for edge detection in MTF tests, the selection of scale parameters and threshold parameters of non-maximum suppression in the Gaussian filter is essential. Meanwhile, the Canny operator is improved by employing an adaptive filter function and maximum inter-class variance method. After the edge detection of the image is completed, the image should be digitized to obtain the ESF curve. Given the need for FFT after obtaining ESF data, an ESF processing method based on a non-uniform fast Fourier transform is proposed.

As shown in Figs. 14 and 15, the MTF detection of the three primary color components is carried out separately, and the color image test results are obtained by weighting according to the calibration results of the light source. This solves the problem that the image quality is further degraded due to sampling rate and interpolation, and the spatial frequency may be cut off in advance. The Canny operator is improved by adopting adaptive Gaussian filter amplification and the Otsu algorithm (maximum inter-class difference method), and the adaptive adjustment of scale parameters and threshold parameters is realized, with the problem of noise affecting edge fitting accuracy solved. Based on the edge extension function processing method of non-uniform fast Fourier transform is a fast algorithm for FFT in non-uniform space, which can both obtain the calculation results quickly and retain the calculation accuracy. As a result, the problem of insufficient homogenization processing accuracy is solved. As shown in Fig. 16 and Table 2, the improved edge method has higher detection accuracy.

To address the difficult MTF detection of Bayer filter color space cameras, we realize the high-precision detection of full band MTF by improving the calculation algorithm and detection method. Firstly, the reason for image quality degradation of Bayer filter color spatial cameras is analyzed theoretically, and an MTF measurement method for each color component of Mono image output by color cameras is proposed, with the MTF detection method synthesized by employing three primary color weight factor calibrations. Then, to improve the measurement accuracy of the knife-edge method, we improve the two steps of edge detection and ESF processing. Simulation results show that the improved knife-edge algorithm is more accurate than the ISO 12233 method. Finally, the MTF measurement of a Bayer filter color space camera is completed. The experimental results indicate that the proposed solution can solve the difficult problem of Bayer filter color space cameras in MTF measurement.