| Wood are widely used in the social production of industry and agriculture and so on, but due to external factors they often have some defects in its natural growth process, and these defects is one of the important factors that affect the wood quality assessment.Therefore, to detect the defects before the processing of the wood is a very important step.At present of our country of wood defects detection technology is not perfect. Although in recent years, researchers have proposed some detection technologies, such as based on C-V model level set segmentation method and based on GAC model segmentation method.But C-V level set model has slowly curve evolution, poor anti noise ability; GAC model is not suitable for optimization of multiple target segmentation, can only deal with a convex optimization problem, easy to produce boundary leakage and do not have self-adaptability.So the efficiency and effect of these detection techniques are not satisfactory, and need to be further improved.And for the image sample collection and transmission of wood defects detection, we have been based on the Nyquist-Shannon theorem, namely:if we want to reconstruct the original signal accurately,then the signal sampling rate must be more than two times the highest frequency of the signal. This requirement will inevitably lead to a lot of data redundancy and waste of computer hardware storage resources.For the problems existed in the wood defect image detection technology, this paper based on the traditional C-V model, GAC model and ROF model, according to the characteristics and deficiencies of these models, these models were integration and improved. This paper puts forward a recognition algorithm for wood defect images which based on the multi-model fusion.The new method proposed in this paper by introducing edge detection function and TV norm in the energy functional to achieve the unity of the three models within the same minimization framework. The image gradient and homogeneous region information are used to detect the target edge of the defect image at the same time, realize the complementary model performance.And then introduce theory of compressed sensing, signal sampling and compression process completed in one step, By developing the signal sparsity, random sampled the signal to get discrete signal samples under the conditions of far less than the Nyquist-shannon sampling standard. This would greatly reduce the redundancy of the sampling process and reduce the burden of the computer hardware storage. At last, the sparse representation of the original image is accurately estimated, and then the nonlinear reconstruction algorithm is used to reconstruct the original signal accurately to realize the image denoising and improve the noise immunity of the algorithm. In the process, the paper also improves the compression sampling matching pursuit CoSaMP reconstruction algorithm, and use matrix block principle to achieve the reconstruction accuracy and reconstruction time reduction.Through the experimental results show that the multi model fusion wood defect image detection technology does not depend on the choice of an initial contour, is not limited by the defect type, shape, number and gray image is uniform and other factors, active contour can stably stay in edge of the object to be measured to avoid edge leakage problems, and compared to C-V model and GAC model reduces the identification time, have a certain universality. The introduction of compressed sensing theory, strengthen the robustness of the algorithm, and the experimental results show that the compressed sensing theory has very good effect for image denoising. |
