Studies On Image Processing Methods In 3D-measurement

With the development of 3D-measurement system based on computer vision, the key image processing methods and technology play a great role in this field. In this thesis, starting with basic multi-scale theory, the main and key image methods in 3D-measurement are studied. The main conclusions are as follows:First of all, the principles and methods of the currently popular analysis tools—multi-scale theory are studied systematically and detailedly. And then, classical Fourier transform and wavelet transform in image processing are demonstrated to show its' characteristics, especially, Ridgelet transform that has been introduced by Donoho in recent years is also discussed. Furthermore, the conclusions of basic problems and main targets based on multi-scale image processing are made.Secondly, by the use of the multi-scale tools, a novel denoising method and image restoration algorithm are presented. The suggested denoising method is reached by fitting jointed probability distributions of parent-children wavelet's coefficients based on mixtured Gaussian models and using the bi-variate Bayes Rules. Besides, the comparisons are done detailedly between the suggested restoration method -Markov ForWaRD in complex wavelet domain and classical restoration methods. Results show the suggested method does improve the quality of the distorted images.Corner detection is one of important tasks in 3D-measurement. So classical algorithms are described, corresponding simulation codes are given and their performances are tested. For overcoming the difficulties of the classically differential algorithms, the suggested method—Fourier-Ridgelet Corner detector are presented. According to the simulation results which are gotten by using the calibration palette images, the suggested method improves the robustness and reduces the error ratios in corner detection.Finally, the parameters of camera calibration are retrieved with the help of the above mentioned methods. In summary, the presented image processing methods in this paper can be applied to the 3D-measurement systems and get good performances.