With the gradual development of precision instruments in the direction of automation and intelligence, higher requirements are also put forward for digital microscopes, among which autofocus based on image processing is the most widely used. Taking the microscopic imaging system as the research object, this paper briefly introduces the three important links of focus evaluation function, focus window and focus search algorithm in focus depth method.
01
—
Bre2d_Rob Focus on the evaluation function
The Brenner2d function increases the gradient calculation in the y direction, and the evaluation effect is better for images with sharp gray changes in the y direction, and it is also more in line with the visual judgment of the human eye. In order to compare the changes of focus sensitivity of the Brenner function before and after the improvement, some microscopic images from blur to clear to blurry were collected, among which the positive focus image is shown in Figure 1(a), and Figure 1(b) is the focus evaluation curve of the Brenner function and the Brenner2d function.
Figure 1 Positive focus image and focus evaluation curve (a).Positive focus image (b).Focus evaluation curve
Considering that the micro-nano structure has multiple edge directions, the Brenner2d function and the Roberts function can be combined to propose a new focus evaluation function, the Brenner2d_Roberts function (hereinafter referred to as the Bre2d_Rob function), which is expressed as follows:
It is known from Equation (3.33) that the Bre2d_Rob function extracts the gray gradient information of the microscopic image from multiple directions, and when the edge direction of the image changes, there will always be a dominant gray gradient direction in the function, so as to adapt to the gray gradient change in different directions and have better focusing stability.
02
—
Variable step focus window
According to the experimental results and several experiments, each focus window selection method has some problems:
Figure 2 Illustration of the central window taking method
(2) Although the multi-area window taking method has a certain adaptability to the target offset and makes up for the shortcomings of the central window taking method, it also introduces too much background information, and still cannot avoid the situation that the imaging subject deviates from the focus window.
Figure 3 Schematic diagram of multi-area focus window selection method (a).inverted T-shaped window method (b).golden section window method
After comprehensively considering the advantages and disadvantages of the above focusing window selection, some scholars proposed a variable-step focusing window selection method, which combines the advantages of non-uniform sampling window taking method and central window taking method, which not only has the ability to retain global detail information of non-uniform sampling window taking method, but also has the advantages of simple calculation of central window taking method. The window taking method divides the entire image into three areas: the central area, the middle area and the edge area.
Figure 4 Schematic diagram of image area division by changing step length window method
This method adds the calculation of edge details on the basis of the central window method, and has certain adaptability to the situation that the image deviates from the central area.
03
—
Improve the traditional mountain climbing search method
The traditional mountaineering search method only judges the search direction by comparing two images, while the actual focus evaluation curve is often not as strict and monotonous as the ideal curve, which makes the mountaineering search method easy to fall into local extreme points, resulting in focus failure.
Figure 5 Flow chart of mountain climbing search method combined with rough and precise combination
This paper takes digital microscope as the research object, and shares with you the microscopic autofocus technology based on image processing.
References:
1.Helstrom,Carl W. Image Restoration By the Method of Least Squares[J]. Josa/57/3/josa Pdf, 1967, 57(3): 0-297.
2.Stites DG. Automatic Focus Sensing and Control of Optical Reconnaissance Sensors[J]. Proceedings of Spie the International Society for Optical Engineering, 1976, 7
*The academic content of the article is from the Institute of Optoelectronic Technology, Chinese Academy of Sciences, if necessary, you can download it by yourself.
Select a method
(1) The premise of the application of the central window method is that the imaging subject is located in the center of the image, and when the target is off-centered, the focusing performance decreases sharply.
To this end, the traditional mountaineering search method is improved, and the focus search is divided into two stages: rough search and fine search, and the focus search direction is judged by three images in the fine search stage, which reduces the influence of local extremes and improves the anti-interference of the autofocus system. The overall autofocus scheme was designed, and an autofocus system with a PC as the image processing unit was built.
