Wood Surface Defect Image Segmentation Based On2D Entropy Method

Our country is lack of forest resources. It becomes a topic for forestry science and technology persons to improve the utilization rate of timber. Therefore, in the process of wood processing, identify and picking out wood defect has important practical significance. This article with slipknot, fast knot, bug three common wood surface defects as samples. Using threshold segmentation method and differential operator to segmentalize the wood surface defects in gray space and HSI space.First of all, preprocessing the wood surface defects images. Using the mean filter, median filter and wiener adaptive filter to filter the wood surface defects images. The experimental results show that using the wiener adaptive filter to deal with the noise of wood surface defect images is better than others. In order to highlight the details information of interested area of the wood surface defect images, we equalize its histogram and transform its gray level.Using differential operator to segmentalize the wood surface defects in gray space. Respectively using Roberts operator, Sobel operator, Prewitt operator, Log operator and Canny operator. The results show that using differential operator method can not form a coherent description defect area, and is influenced by wood texture strongly. Using threshold segmentation method to segmentalize the wood surface defects in gray space. On the basis of the traditional threshold segmentation method, introducing a method named2D entropy image segmentation. This method uses image’s gray level and gradient value. So we introduce how to build the gray level-gradient co-occurrence matrix. Compared with the traditional threshold segmentation methods which including the iterative method, the Ostu method and the one-dimension maximum entropy method, we find that using the two-dimensional entropy segmentation method is better than the others.For normal wood surface defect image, it is always black or light yellow in hue. Its color saturation is low. The impact on the color will be very small, can be ignored. So we only deal with the H component and I component. After segmenting the two components with threshold segmentation methods we fuse them together.Finally, we post-process the images which are segmented with threshold segmentation methods in gray space and in HSI space. By adopting the method of mathematical morphology for image segmentation expansion, area filling, morphological filtering, edge detection and mask operation. The experimental results show that based on two-dimensional entropy in HSI space of wood surface defect images segmentation result is better, more complete retained the timber surface defect information, and there is not less segmentation or more segmentation phenomenon.