Research On Stereo Matching Technology For The Stereo Microscope

Micro stereo vision system has been widely used in many aspects, such as biology,medicine, industry, operating methods, but the processing technology for macro scenario isdifficult to be directly used in micro stereo vision system, and therefore we urgently need todevelop technology and method which can be suitable for processing of the tiny objects. Themicroscope is an important tool to obtain the tiny objects. Stereo microscope can obtain twochannel stereo images under the same scene at the same time; we can obtain three-dimensionalinformation of objects through the two stereo images, which is used for three-dimensionalreconstruction and three-dimensional measurement. This article mainly studied in the aspect ofauto-focusing in the process of acquisition of the micro stereo vision system, color correction,stereo matching in the process of three-dimensional reconstruction and three-dimensionalmeasurement.In stereo matching process, the problem of image acquisition must first be solved, comparedwith the traditional monocular camera, the depth field of the stereo microscope is very small. Thesmall changes in the process of acquisition can lead to the sharpness of the images existence verybig difference. An auto-focusing technology in the process of acquisition is the one of the keyfactors to solve the image resolution. This thesis proposed an auto-focusing method based onedge detection in the process of study, the edge detection was firstly used to the image sequence,and then the edge detection result can be obtained. For the edge detection results, the globalthreshold was used to eliminate the false edge because of the noise impact, last, we usingfocusing evaluation function in the edge detection of images to find the focus position. Theproposed method of auto-focusing algorithm firstly using local small window for noisesuppression in the process of edge detection, and then using global threshold to eliminate thefalse edge, on the edge location to evaluate the sharpness of the image. The experimental resultsprove that the auto-focusing algorithm can very well to resist the interference of noise and alsoachieve high sensitivity.The differences of CMOS manufacturing existed in the stereo microscope, and the CMOS sensor is sensitive to the light environment in the process of imaging, it is result in the capturedimages exist big difference. In order to solve this problem this thesis proposed two types of colorcorrection methods: The first color correction method is based on salient region detection. In thefirst method we mainly considering the foreground and background regions in the collectedimages of the stereo microscope can be distinguished and segmented, we using colortransmission in the corresponding foreground and background regions, which can effectivelyimprove the accuracy of color correction results. In the process of implementation, we first usecolor contrast method to obtain the salient regions, and then using the GrabCut method to get thesegmentation results for the salient regions. In the corresponding regions between the referenceimage and the distorted image through the cumulative histogram matching to obtain the colortransfer function for the three different channels; The second color correction method is mainlyachieve in pixel level. The homograph matrix can be well used to build the corresponding pointsin the stereo images, and then obtain the position of the corresponding point coordinates throughthe homograph matrix. We use SIFT operator as feature points extraction operator, which hasvery good stability to image scaling, rotation, illumination changes, and using the proposedoptimization of feature points matching algorithm to achieve the point matching. At the sametime, we proposed a new color transfer method to achieve color correction, which is based on theprinciples of color similarity and distance proximity.Stereo matching is an important research direction in the micro stereo vision system. Thereare two kinds of different methods of stereo matching, the local stereo match method and theglobal stereo match method. The methods only uses the color information to achieve stereomatching is suitable for the stereo images, which the difference between each other is small andthe geometric distortion is not evident. The color of stereo images, which captured by stereomicroscope are obvious and the deviation of the geometric position is also obvious. The methodonly using the color and gradient information to achieve stereo matching has failed. This thesisproposed a stereo matching method, which is suitable for stereo microscope. Firstly, we using alocal feature extraction operator to obtain multi-dimensional vector for each point in the stereoimage, each of the multi-dimensional feature vector contains the reference relationship betweenthe current point and the neighborhood points, similar to the SIFT feature points, the DAISYoperator is stable to scaling, rotation, illumination changes. Firstly, we using the direct matrix toobtain the corresponding points position, and then by the multi-dimensional feature vectorthrough euclidean distance to calculate the difference between the corresponding points, thecalculation result then be used to establish the initial matching cost function. Finally, we used thedisparity optimization method to obtain the initial disparity for each point in the stereo images. In order to optimize the disparity results, we combined with the image segmentation method ofMeanShift to achieve image segmentation through calculating plane fitting parameters for eachsegmented regions to correct the error disparity.