Research On Pearl Classfiction Technology Based On Monocular Multiview Machine Vision

China has the largest pearl production in the world, but as the largest country of pearl industry, the way for pearl grading in China is mainly by handcraft. This grading method not only requires much time and effort, but also is susceptible by workers’ subjective factors. In order to solve the issues mentioned above, a pearl grading system based on monocular multiview machine vision is designed and implemented in this thesis.The main research work and results of this thesis are as follows:1.According to pearl’s non-planar characteristic, a monocular multi-view detection device is designed. The device could get the entire surface of the pearl by one shooting of one camera. Using this device, the detected pearl’s entire surface could be achieved quickly, and all images of the pearl have the uniform color system.2.According to the characteristics of the visual detection device designed in this thesis, the corresponding algorithm of image preprocessing is researched to accomplish the image filtering, foreground extraction, location and normalization.3.In order to accomplish the detection of pearl’s appearance, a dynamic linear weighted image mosaic algorithm is designed to acquire pearl’s panoramic image. The algorithm gives dynamic weight in different positions according to the regularity of the overlap region’s RGB value in the two images. Experimental results show that the algorithm proposed in this thesis could solve the problem of splicing gap and obvious color difference appeared in traditional weighted average algorithm.4.According to the national criterion of classification, the pearl’s shape, size, color, luster and blemish indexes are detected by machine vision. The shape detection mainly has two steps: Firstly, edge detection is used to get the contour information of the pearl. Secondly, the Fourier Series is achieved by Fourier transform of edge points to accomplish the assessment of the pearl shape. The H value and S value in HSI color model are used to realize the detection of pearl color. The I value in HSI color model is used to accomplish the detection of pearl luster. The Laplace of Gaussian algorithm is used to realize the blemish detection.5.The application system is designed and implemented. In this thesis, the system consists of the hardware device and the software system. The design of hardware device mainly includes the design of mirror cavity and lighting source. It is used to realize multi-view acquisition of pearl image. The software system mainly includes preprocessing module, the module of color and luster detection, the module of size and shape detection, the module of blemish detection. It is used to accomplish the visual detection of pearl indexes.This thesis makes some attempts and exploration in the application of monocular multiview machine vision to the classification of pearls, and gains some achievements which build a foundation for the future commercial grading system of pearls. The experimental results show that this detection method using machine vision could achieve high accuracy and real-time performance for the appearance index of the pearl which has the obvious characteristics.