Application And Analysis Of Texture Features On Tangential Section Image Of Moso Bamboo Strip

The computer vision system was established with industrial camera, standard illuminant and microcomputer. To capture the tangential section image of bamboo (Phyllostachys pubescens), an image capture program was written with VC++.net. Captured the images of bamboo knot area and internode area on bamboo tangential, and denoised the images, the texture features of the images were analyzed in the airspace or frequency domain, based on Image texture analysis method including mathematical morphology, gray level co-occurrence matrix and fast Fourier transform. The difference of image texture features between bamboo knot area and internode area were compared to screen the texture characteristic parameters. And then, a BP neural network was built by using the optimum texture feature vector as input parameters and whether an image contained bamboo knot as output parameters. After training, the BP neural network can be used to accurately identify the bamboo knot area and internode area.Analyzing the texture composition of tangential section on bamboo, the length, width, area of vascular bundle and rectangle degrees of vascular bundle were chosen based on mathematical morphology to describe the texture features on tangential section of moso bamboo. The texture features extraction program that was written with VC++.net was conducted to extract the texture feature from lots of knot area images and internode area images. Analyzing and comparing the texture feature between image of knot area and internode area on tangential section of bamboo, The max length of vascular bundle has significant difference between knot area and internode area, it can be used as a characteristic parameter to judge whether is the image of knot area or internode area.To describe texture characteristic on tangential section of bamboo based on gray level co-occurrence matrix, some features were choosen which inculuded meanvalue, variance, moments deficit, contrast, non-similarity, entropy, angular second moment(energy) and correlation. Writing the program to generate the gray level co-occurrence matrix and calculate texture feature parameters with different structure factors were optimized, which comprised gray level,direction and distance. Analyzing and comparing the texture features distinctive between knot images and internode images, some texture features were screened that had significant difference between knot images and internode images, which included energy, moment deficit, contrast, non-similarity and entropy based the gray level cooccurrence matrix.Taking advantage of fast Fourier transform to pocess digital image and describing the texture feature, some texture feature parameters in frequency domain were introduced to describe the texture feature on tangential section of bamboo. The parameters describing the texture feature in the frequency domain included mean, variance, entropy and energy ratio on rectangular ring of amplitude spectrum. The program of fast Fourier transformation and calculating the texture parameters was written with VC++.net. Analyzing and comparing difference of the feature parameters between knot images and interrode images,5texture features had obvious difference which included variance, and energy ratio of the third, seventh, eighth, ninth rectangular ring between knot images and interrode images.The max length of the vascular bundle, energy, moments deficit, contrast, non-similarity, entropy, variance and energy ratio of the third, seventh, eighth, ninth rectangular ring of frequency spectrum were built a texture feature vector to describe the characteristic of tangential section of bamboo. BP neural network was built with the neural networks toolbox of MATLAB software, input parameter of BP neural networks was the texture feature vector of, and output parameter was whether was knot image. Training the BP neutal network, and capturing lots of knot image and internode image of bamboo tangential setion, the BP neural network was examined, and the results shown that correct rate of BP neural network to distinguish whether was knot image was more than99%.