Study On Auto-focusing Methods Based On Image Processing Technology

Image is an important carrier of information transfer and how to acquire clear and high-quality images has always been direction and objective of exploration. Besides, image quality has direct relation with subsequent image processing and application. Automatic focusing technology, as one of key technologies of optical imaging system, has been widely applied to precise instruments like medical microscope and scanner as well as in fields such as satellite navigation, space exploration, robot vision and automatic monitoring, etc.Different from traditional focusing technology, automatic focusing technology based on image processing relies on acquired image information without need of additional photoelectric inspection equipment and it’s divided into two classes according to image processing mode: depth from defocus and depth from focus method. This paper mainly conducts in-depth study of depth from focus method. Focusing process based on depth from focus method is a closed-loop control process which integrates light, machine and electricity. It mainly includes three important links: construction of focusing window, selection of focusing evaluation function as well as feedback control of focusing motor by search algorithm. Based on study of developmental status of automatic focusing technology both abroad and home, this paper has conducted a series of researches by combing practical application directing at these three important links.Firstly, this paper established out-of-focus model of optical imaging system and introduced point spread function and optical transfer function under out-of-focus status. It also laid theoretical foundation of automatic focusing technology by analyzing essential reason of image blurring caused by out of focus.Then, selection of focusing evaluation algorithm is a crucial step in focusing process. Out of focus will results in loss of image high-frequency components which is manifested by blurring image edge and details as well as decreasing image quality in spatial domain. This paper analyzed common focusing evaluation functions based on image gradient, frequency domain, information entropy and statistics. Image power spectrum based on Fourier transform has been widely applied because of its characteristic—scene invariance. Wavelet analysis has overcome the defect of single resolution of Fourier and has characteristic of multi-resolution analysis. Wavelet power spectrum function is an improvement of power spectrum function based on Fourier transform. By combining low-pass filtering characteristic of human visual system(HVS), weighting processing is conducted on wavelet power spectrum, which has made evaluation results accord with visual perception characteristic of human eyes more. In order to verify effectiveness and universality of the algorithm, relevant experiment was conducted on LIVE database. In the meantime, directing at characteristics of large-scale optical measurement device, this paper proposed an evaluation algorithm based on self-adaptive threshold segmentation and improved SML sharpness which added gradient values in direction of two opposite angles based on original SML function. The proposed algorithm took the maximum value as spatial filtering results of this pixel. This algorithm can effectively overcome the influence of background noise and equipment jitter caused by atmospheric turbulence with high stability and sensitivity.Subsequently, this paper studied the influence of size and position of focusing window on focusing evaluation curve. As observation targets of optical measurement device has the uncertainty of position and size, oversize window will cause interference of background information and undersize window will lead to target loss. Base on analysis and comparison of limitation of selection method of traditional focusing window, this paper proposed a window construction method based on multi-scale pulse cosine transform. This method simulates human visual attention mechanism and extracts area-of-interest of image. Besides, it can make optical measurement device adapt to position variation of imaging subject under any focusing status and establishes effective focusing window which improves the timeliness of the system.Afterwards, a search strategy was designed which is the final realization of focusing process. There exist some problems of several common search algorithms in practical application. Thus, an improved common hill-climbing method with high feasibility was put forward. Improved hill-climbing method was taken as basic search algorithm of focusing system while wavelet power spectrum value of image as auxiliary means to comprehensively judge the focusing search direction and property of extreme point; Self-adaptive step length was adopted to improve searching velocity. The proposed search strategy can rapidly and accurately acquire optimal focusing position and avoid interference of local extreme points with large scope.Finally, according to proposed focusing strategy, this paper designed embedded automatic focusing system with core being DSP chip TMS320C6678 and Virtex-II series of XC2V3000, and it conducted a large quantity of tests and experiments in aspects of focusing precision, stability and timeliness. Experimental results show that automatic focusing method proposed in this paper is feasible with high engineering application values.