The Research Of CCD Natural Image De-noise And Segmentation In Machine Vision

Scientific research and statistics show that Vision System assist human acquire more than 3/4 information form environment. To intelligence machines, endow vision ability to them is necessary; therefore a new subject, machine vision was developed. Images de-noise and segmentation technique as the importance portion of Machine Vision, are widely applied in practice and get abundantly regards.This paper is supported by the projects of Vision Coordinate Measure Machine and Optimized Model of Vision Coordinate Measure Machine. It involves the theory and measurement domain applications of the CCD natural image de-noise and segmentation in low level and middle level vision. The main research works include: analysis CCD natural image noise, proposes a noise statistical model of affecting stereo matching precision and designs a de-noise filter, establish the diffusion equations based on scale gradient images and designs multi-scale image segmentation arithmetic, determine the optimal segmentation scale, designs an active pursuit segmentation method. The main achievement of the research works can be summarized as follow:1. Classified the random noise of CCD natural images into three categories: measuring noise, illumination noise and noise coming from parallax according to experimentation, qualitative and quantitative analyze them. Proposed a straightforward random noise statistical model based on the stereo matching characteristic in binocular systems. Therefore, an algorithm to reduce noise while preserving useful information is designed. Experimental results show that the algorithm has the advantage of robust and can make 3D measuring precision of binocular systems better than 1/4 pixel in practical applications.2. Established four diffusion equations based on scale gradient image by feedback and give their physics interpretation respectively. The simulation results consistent with the theoretic analysis. Apply the nonlinear forward diffusion equationas multi-scale natural image segmentation method. It can remove the detail component of images and preserve the primary part.3. Designed two algorithms to determine the optimal segmentation scale. One based on the cross-information, another based on sparse decomposition parameters. All in all, the two algorithms ensure the image in semi-equilibrium state after segmentation, but the simulation results prove latter method more effective than the first one.4. Compressed gray range to engender active pursuit images. It simulates the biology vision system pursuit targets via moving eyeball in principle. Established the neural network model of active pursuit image segmentation shows that the segmentation method accords with the Multiple Synchronization Integration hypothesis of vision information transmission.5. The research fruits of CCD natural image de-noise and segmentation have been applied to vision measurement. They include: the traffic accident scene surveillance system and artillery automate butt measurement system.