Research On Segmentation Algorithm Of Industrial CT Image Using Improved CV Model

Industrial computed tomography (ICT) technology has been widely used to locate inner defects and measure geometry including area and volume. However, it’s difficult to determine the properties above only by the2D slice images. As a result, the segmentation of the image directly using the CT volume data is getting more and more attention. Based on active contour, CV model has been perfectly applied in the fields for segmentation, however, we can hardly get satisfactory results when dealing with images with different kinds of noise. To address the problems above, we propose several improved CV models. Then we extend the models to three dimensional and apply them for segmentation of industrial CT volume data.As for two kinds of typical noises, white Gaussian noise and salt-and-pepper noise, there are two corresponding filtering algorithms, median filter and Gaussian filter. Both filters share a common problem that how to choose the parameters. Specifically, as for median filter, the choice of window size determines the effects of the filter. When the filter window is bigger, the capability of suppressing noise is stronger, but the details and edges of the image are hardly preserved, as a result there is too much information neglected. On the contrary, when the filter window is smaller, too much noise cannot be well suppressed. At present, there are many improved median filter, for example, AMF filter which has strong capability of suppressing noise, and IWMF filter which has strong ability to preserve details of the image.For Gaussian noise, the choice of variance determines the effects of the filter. Similar with the case of median filter, when the variance is bigger, the capability of suppressing noise is stronger, but the details are hardly preserved. On the contrary, if the filter window is smaller, noise can’t be well suppressed.In this paper, we present a new improved3D CV model to deal with the problem with the help of boundary information during the evolution of CV model. We consider the distance between the pixel and the boundary curve during the evolution of CV model and use it to help us improve the traditional filter. Specifically, as for salt and pepper noise, we propose a brand new adaptive median filtering method with the help of boundary information; we choose the median filter based on the inverse of the weight (IWMF) when a pixel is adjacent to the boundary and the adaptive median filter (AMF) method otherwise. As for Gaussian noise, we propose a brand new Gaussian filtering model which the variance can be determined adaptively according to the evolution process of the CV model. Similar with median filter case, when we use distance to define the variance of Gaussian filter, we choose a relatively smaller value of the variance when the pixel is adjacent to the boundary and a relatively larger value otherwise. Then we propose an adaptive robust CV model based on statistics algorithm. At last we extend the proposed improved CV model to three dimensional cases and apply it to the segmentation of industrial CT volume data.