Auto-focus Accuracy And Stability Of The Evaluation Function

Auto-focusing technology is a crucial technology in the robot vision and digital video frequency system. At the same time it is also an absolutely necessary crucial technology in modern optics imaging system. Now the technology has been used abroad in the camera, the vidicon, the microscope, the scanner and so on. With rapid development of science and technology, people think much more of auto-focusing issue. Many researches have been done on auto-focusing algorithm, however, there are still many problems remain unsolved in auto-focusing algorithm, such as, the precision, stability and anti-noise of the algorithms. The perfect clarity-evaluation-function should have the impartial and the single apex. As to images with having different contrast, the algorithms have good stability, at the same time, as to images with noise, the algorithms have good anti-noise.Based on Fourier optical theory, PSF (point spread function) and OTF(optical transfer function) are introduced. Then the mechanism why digital image is focus in optics imaging system is also analyzed. Thus the arithmetic is naturally achieved. Then we researched the existing auto-focusing algorithms thoroughly, found their deficiencies and the resolvent.By the analysis of the OTF, the lens when the image is off the focus can be as the low-pass, and the high frequency lose. After we research the distributing of image gradient, we found that the edge pixels had some characteristics: (a) The gradient value of edge pixels is bigger than the gradient value of non-edge pixels. (b) The number of edge pixels is small to all pixels in the image, and the number is smaller than 2%. So we can use some threshold to reduce the influence of non-edge pixels, make the algorithms have better precision and the image is focused. In the paper, the RGA is introduced for the first time, what is more, the mathematical model has been established individually.In addition, in the research, we find the classical algorithms are sensitive to noise, these algorithms are very good without the noise, but these algorithms will be useless when the image is polluted by the noise. In order to solve the problem, the ADPA is introduced.At the same time, in the research, we find some classical algorithms based on gradient have some problems. Firstly, when these algorithms complete the gradient, we select the certain gradient direction. However, different images have different gradient direction, and the true gradient may be different from the certain gradient, thus there are some errors. In addition, the image usually be polluted by the noise, but the algorithms based on the gradient is sensitive to the noise. These algorithms are very good without the noise, but these algorithms will be useless when the image is polluted by the noise. In order to solve the problems, the MGT is introduced.The computer simulation result shows that, the clarity-evaluation-function of out-of-focus blurred image based on RGA, ADPA and MGT has the characteristics of non-deflection, single peak value, high sensitivity and SNR. And the speed and veracity of focus are improved and the data of image is reduced. As for to image with noise, the APDA and MGT also have good character and anti-noise.To sum up, RGT, ADPA and MGT arithmetic, which are proposed in this paper, all are effective auto-focusing ways.