Study On Auto-focusing Method Using Image Technology

As a key technology in the optical imaging system, the autofocusing technologyhas been successfully used in the space optical remote sensing system, the rangeoptical measurement system, the automatic monitoring system, the microscopyanalysis system, a variety of digital cameras system, and so on. In an imagingsystem, accurate focusing of the optical lens is the most important issue that needs tobe solved before all other functions. The final results of autofocusing directly impacton the imaging quality and the effectiveness of subsequent imaging processing andits applications. In order to solve the problems of the existing autofocusing methods,two kinds of autofocusing method, including depth from defocus (DFD) and depthfrom focus (DFF) based on image technology, are studied. Defocus from defocusalgorithm based on image restoration, defocus from defocus algorithm based onestimating blur amount, as well as the problems of the depth from focus method,such as focus measure function modeling, the smart selection of focus window, thedesign of focus searching strategy, are also deeply studied. Moreover, theexperimental platform for autofocusing system has been designed. The maincontributions of this dissertation can be summarized as follows:1. Two kinds of point spread function models, including the disc defocus modeland the Gaussian defocus model, commonly used in the defocus blurred imagerestoration algorithms are analyzed. Since these two models can not accurately approximate to the actual defocus point spread function, leading to the defocusedimage restoration algorithms are incapable of achieving the desired results. Theshortcoming of the non-negativity and support constraints recursive inverse filtering(NAS-RIF) algorithm is analyzed, and a NAS-RIF blind algorithm for defocusblurred image based on lifting wavelet transform is proposed. This algorithm notonly can restrain the noise effectively, but also can restore the detail edges of thedefocus images. In other words, the best restoration of defocus images is acquiredby using a single blurred image.2. The defocus model of the imaging system is analyzed. The presentation of theautofocusing algorithm based on DFD is provided. Under the condition of onlychanging lens position of imaging system, a formula to compute the depth fromdefocus of target with respect to the blur difference is derived according to thedefocus model of imaging system. A Spatial-Domain Convolution/DeconvolutionTransform Method (STM) is applied to estimate the blur difference between the twodifferent defocus images, and then the depth information of target is computed withthe derived formula. Consequently, the lens could be adjusted to accomplishautofocusing. This method has no excessive requirements on the imaging systemand imaging target, so it is applicable to any target.3. The study of the existing focus measure functions based on image gradient,frequency domain, informatics, statistics and wavelet analysis are performed. On thebasis of the mechanism of biological vision, with the feature of wavelet transformmatched to the multi-channel characteristic of the human vision system (HVS),combined with a band-pass characteristic of contrast sensitivity function, a weightedwavelet focus measure function is proposed based on HVS. This function not onlycan objectively and accurately evaluate the defocused images and the focusedimages in the process of autofocusing, but also can satisfy the perceptual property ofhuman eye. The width of steep part of focus measure curve, the steepness, the ratioof sharpness, and the factor of local extreme and the sensitivity are selected to beevaluation indexes. And then the characteristic of the weighted wavelet focus measure function based on HVS is analyzed by comparing with several other focusmeasure functions.4. The influence on the autofocusing effect of the focus window selection isstudied. The influence of focus window size and focus window position toautofocusing are analyzed, respectively. The limitations of the traditional focuswindow selection methods, only selecting fixed focus areas and not well-suited forthe change of the target position, are analyzed. For the case of the target position isnot fixed, a focus window selection method is presented based on self-adaptivegenetic algorithm (GA). This method can effectively reduce the focusing failure dueto the target not located in the center of the imaging field.5. The existing issues of the current focus search algorithms in practicalengineering applications are analyzed. In order to overcome the disadvantage of lowsearch efficiency of the mountain climbing search algorithm, an improved searchingalgorithm is presented based on the self-organizing map (SOM) neural network.This algorithm can significantly accelerate the search speed because it can avoidredundant movement of the motor around the best focused lens position.6. Aiming at the proposed autofocusing method, the experimental platform forautofocusing system has been designed, and its implementation is also proposed.Subsequently, series of autofocusing experiments are carried out. According to theexperimental results of the test, the detailed analyses of the accuracy, real-time andstability of autofocusing are made. The experimental results demonstrate that theproposed autofocusing method has a high feasibility and can provide a solution forpractical applications.