Research On Auto-focus And Image Stitching Method Based On Micro-vision

Microscope,as an important optical instrument for human to observe the microscopic world,plays an irreplaceable role.In order to obtain clear microscopic images,people adjust the focus of the microscope manually.This pure manual method is inefficient and brings the error.Auto-focus technology can eliminate the artificial error,and can accurately find the best focus position by program.However,auto-focus technology has the problem of tedious search steps,over relying on evaluation function values and vulnerability to local extremums.In addition,due to the narrow field of micro-vision,how to obtain clear panoramas has become a hot issue in the current research.In this paper,the auto-focus and image stitching of microscopic images are mainly studied based on the automatic microscopic operation system.The main research contents of this paper are as follows:1.Based on the existing hardware,the software of the automatic microscopic operating system is redeveloped.With the use of the development kits provided by hardware manufacturers and the programming software,an upper machine software which can realize the function of microscopic images acquisition is designed.The software can realize a series of functions such as automatic scanning,auto-focus,image preprocessing and image storage for microscopic images.It basically realizes the basic functions required by a microscopic operation software.2.As for auto-focus methods,the focus evaluation function and focus searching algorithm commonly used in auto-focus are studied in this paper.For the problem that the focusing evaluation function is not suitable for different kinds of microscopic images,a new focusing evaluation function is proposed by weighting the gradient energy evaluation function in time domain and the discrete wavelet transform function in frequency domain.For the problem that the classical mountain climbing searching algorithm has many search steps and is easily affected by local extremum points,this paper designed a method of auto-focus microscopic images based on Self-Organizing Map neural network,which inputs training samples into the neural network for training,and accelerates training with Particle Swarm Optimization algorithm.In the training process,the trained neural network can classify and recognize the measured samples,predict the best focus position,and then fine-tune it by mountain climbing searching algorithm until it finds the best focus position to complete auto-focus.The experimental results show that the auto-focus method effectively solves the problems of slow auto-focus efficiency and poor anti-noise performance.3.As for image stitching methods,several classical image feature extraction algorithms are analyzed and verified by experiments.This paper mainly studied the Scale-Invariant Feature Transform algorithm.For the problems of large amount of calculation,long operation time and mismatching of SIFT algorithm,this paper improves the feature descriptor of SIFT algorithm,reduces the dimension of the descriptor by transforming weighted method,and combines Principal Component Analysis algorithm to reduce the dimension of the improved SIFT algorithm twice.The experimental results show that the improved algorithm still has scale,illumination and rotation invariance,and feature extraction information is complete,which effectively reduces the mismatch rate,reduces the operation time and improves the efficiency of image stitching.Finally,we use frame-to-frame image mosaic method,and use weighted average method to do image stitching experiments on the collected focused microscopic images.Finally,a smooth panoramic microscopic image without obvious gaps was obtained.