Autofocus Algorithm Research And Verification Platforms And Design

Autofocus is a key technique in digital imaging system, computer vision and microscope instruments etc. Traditional autofocus module can not meet the requirement of integrating into digital imaging devices. Many researches have been done on autofocus algorithm, especially in FMF(focus measurement function).However, there are still many problems remain unsolved in autofocus algorithm, such as how to select focus region and improve the efficiency of peak searching.The paper focuses on the image processing autofocus algorithm applied in digital camera. The development of algorithm verifying platform based on computer has been referred. Autofocus algorithm consists of three modules: focus region selection, focus measurement function and peak searching algorithm. FMF evaluates the defocus degree of interesting region image. Focus measurement is a kind of reference image quality evaluation. Autofocus is achieved by peak searching algorithm according to the trend of FMF curve, and feedback to the computer until gets to the best imaging position.According to the optical transfer theory, the lens system is equivalent to a low pass filter when defocused imaging, and the high frequent components of image is suppressed. Therefore, traditional FMFs select high components as the measurement. In order to improve the sharpness and antinoise ability of the FMF curve, The paper proposes image edge energy function and MRA(wavelet Multi-Resolution Analysis) function. The Gaussian white noise influence on FMF is also given in the part. Experiment based on autofocus image sequence shows that the edge energy function and the MRA function gain better performance than traditional ones in both sharpness and antinoise ability. Region FMF curve experiments shows the principle of focus region selection: Selecting high SNR image region to ensure the sharpness of FMF curve, Selecting larger windows as possible to ensure the stablization of the FMF curve. The FMF curve of foreground region is different from the background curve according to imaging theory. The FMF curve may be multi-peak and results in focus failure, if focus region contains excessive background image region. Interesting region is mostly the foreground image, Focus region selection has to estimate the foreground region for image segmentation algorithm is not practical. Advantages and shortcomings of centre region selection, multi-region selection, variant sampling region selection and skin...