Research On Autofocus Technique In Digital Holography With GPU

Digital holography has many advantages such as high speed, noncontact, real-time three-dimensional measurement, so it is widely used in micro-device detection, particle field analysis, biological microscopic observation and optical devices detection. Although digital holography can easily reconstructed object using a computer, in order to get the best reconstructed image, we need to use AutoFocus technology to find the best distance for image reconstruction. Two measurement targets must be taken into consideration: one is to find the accurate distance for the best image reconstruction, the other is to autofocus as fast as possible.Based on the research of focus evaluation function and AutoFocus hologram recording,we propose a new focus evaluation function using db4 Wavelet, which can not only accurately determine the best distance with strong noise immunity but also use the advantage of GPU to optimize and accelerate the realization speed of Autofocusing algorithm. This article divides into three parts as follows:Firstly, this paper analyzes two structures of digital holographic recording and three algorithms of images reconstruction, and we adopt off-axis holography optical structure and Fresnel transform method in this article. In addition, principles and forms of several frequently used focus evaluation functions, such as gray variance method, gradient square function and Fourier spectrum logarithmic method, have also been presented.Secondly, a new focus evaluation function based on frequency localization of wavelet transform is proposed. The comparison between the new proposal and other conventional focus evaluation functions verifies its superior performance that can accurately find the best reconstruction distance. Through analyzing the advantages and disadvantages of Fibonacci search method, the "blind" climbing search method and traversing search method, their scopes of application are obtained.Lastly, the new autofocus algorithm is loaded to GPU due to its powerful parallel computation to accelerate this algorithm. The calculation speed of autofocus algorithms running on GPU is 10 times faster than running on CPU. An autofocus software based on GPU with a concise user interface is developed to facilitate other users.