Digital Video Signal Processing Of Stereo Microscope

Digital stereo microscope is a microscope with two digital video camera which can acquire signals from left and right light path. Being the product of the combined application of informatics, artificial intelligence and optics, it has great research value and application value. Stereo microscopic measurement is achieved through processing stereo digital microscopic image. It is the core technology of digital microscopic signal processing. This paper aims to study the key technology of digital microscopic signal processing, explore the key technology of measurement, identification and detection, and discuss robustness of these algorithms.(1) Stereo calibration is key to stereo measurement. Due to the existence of shallow depth of field, however, common calibration algorithm cannot work normally for digital microscope calibration. This paper comes up with a calibration method which is based on depth feedback correction. By using the proposed method, digital microscope calibration can be achieved. Based on calibration, this paper explores eliminating mismatching and virtual interpolation during the measurement process, puts the calibration result into practice, and gets a favorable effect. When calibration is performed, pre-calibration is achieved firstly through pre-set video camera parameter and calibration board image within the range of shallow depth of field; then stereo reconstruction is achieved; and finally compensation is given to the calculated depth and compensation factor is obtained. Thus, the final calibration result can be acquired. The experimental result shows that the recovered Z axis error is-4.43% by using calibration parameters; When stereo mismatching pairs are eliminated, the algorithm is as followed. First, left and right image are matched by SIFT feature. Then, reconstructing is performed by using the calibration parameters and matched feature pairs. These reconstructured points in the world are re-imaged in the left and right image again. Final, through analyzing the difference between re-imaged point and source matched point, mismatching feature pairs are eliminated effectively. The experimental result shows that the mismatching feature points can be 100% eliminated and the right matching pairs can be retained by using the proposed method; When the depth of groove in IC chip is measured based on microscopic measurement, virtual interpolation is adopted to increase matching pairs because there is the little matching SIFT feature pairs of IC chip. According reconstructuring the world coordination of the matching feature pairs using the calibration parameters, the depth of IC chip groove is computed. The experimental result shows that the error in three location is 26.637%, 16.733 and 30.468% respectively.(2) The parfocality of continuous zoom microscope is explored. Parfocality features the constant clarity of target or clarity through fine tuning while high magnification is changed to low magnification. This paper predicts the parfocality of zoom microscope through the discrete point testing under different magnification. Its major contribution is to adopt the in-focus judging technology which is based on comprehensive definition functional and achieve reliable judging under different magnification. The experimental result shows that in-focus judgment and in-focus location of the lens can be obtained by using the proposed method, and par-focus curve can be computed. Besides, this paper brings in dark channel model and explores the influence of illumination restraint or inhibition on definition judging to make the judging algorithm more robust. First, dark-channel of source image is computed. Then the original without transmitting attenuation and target area is obtained through dark-channel. Finally, the sharpness of target area of source image is computed. The experimental result shows that the robust is improved whatever sharpness algorithm is chosen such as time domain, frequency domain or transform domain.(3) Microscope can only offer a narrow field of view. To improve this, splicing technology is necessary. Chapter 4 of this paper explores the splicing methods of microscopic image. Through SIFT feature extraction, feature matching, and mismatching elimination, equation of motion of two images that comply with each other can be acquired and image splicing can be achieved. This method is applicable to rigid motion 2D+t image splicing. Sometimes due to the existence of dust, there is grain noise in image. The paper makes the accurate positioning of noise possible through Gaussian-like model and subsequent iteration so as to lay a foundation for the filtering of subsequent noise and image recovery. The experimental result shows that the proposed Gaussian-like model can located grain noise effectively.The science questions in research on stereo microscope are studied systematically in this paper. There are some novel reseach results in the aspects such as stereo calibration, stereo measurement, stereo matching, parfocality acquiring, robust of sharpness computing for illumination, fixed noise location and images mosaic.