木材表面缺陷图像识别的算法的分析研究.doc

PAGE 木材表面缺陷图像识别的算法研究 摘要 随着木材加工业的集约化发展，木材产品的生产量持续大幅度增长。在生产中，对木材表面加工质量高水平的苛求，尤其是一致性的要求，使得传统的人工检测方式已经难以胜任。为此，本论文基于机器视觉理论对木材表面缺陷识别进行了深入研究。 结合数字图像处理技术和支持向量机模式识别技术，本论文研究了木材表面缺陷图像预处理、特征提取、模式识别问题，研究并改进了用于检测木材表面缺陷的定位和识别等图像处理算法。 图像的预处理是检测的第一步，它对图像缺陷特征的正确提取是非常关键的。本文针对传统滤波算法在抑制噪声的同时，也会对图像的边缘及细节有比较大的损害，使图像的边沿及细节变模糊的问题，提出了加权有向平滑滤波算法。并在图像分割上融合了几种分割方法，提出一种改进的基于双正交小波变换的多分辨率图像融合方法和基于融合技术的小波变换和形态学边缘检测算法，优化了分割效果，为后续特征提取打下了很好的基础。 对于木材缺陷的识别，本文从纹理特征（5个灰度共生矩阵参数）和颜色特征（4个颜色矩参数）两个角度来描述缺陷。根据各参数分布情况，选择标准差较小的参数作为分类器输入特征向量；以及采用主分量分析法进行特征提取，降低纹理特征维数，消除模式特征之间的相关性，突出其差异性，满足识别层的输入要求。并采用支持向量机分类器进行缺陷的模式识别，达到较高的识别率。 实验结果证明：根据木材表面缺陷图像的纹理特征和颜色特征，运用数字图像处理技术，来解决木材表面缺陷的分割和识别等问题，是行之有效的途径。 关键词：数字图像处理技术；图像分割；特征提取；支持向量机 Abstract With the development of wood industry, the manufacture of wood products is increasing significantly. The demand of a consistent high-quality surface wood product introduces automatic inspection that cannot be easily satisfied by traditional manual inspection. Based on the theory of computer vision, a research on defect distinguish of the wood surface is made in the paper. Image preprocess, feature extraction and pattern recognition of wood surface defect images are also studied by means of digital image processing technique and pattern recognition technology based on SVM (Support Vector Machines). Image processing algorithms are studied and improved to orientate and recognize wood surface defect. Image preprocess is the first step for detection, which is vital to the correct extraction of the defection feature. In the fact of a traditional filtering algorithm can substantially damage the edges and details of the image and blur the image’s edges and details, a weighted and directional smoothing algorithm is proposed in this paper. Merging several image segmentation method , a improved method of image fusion of multi-resolution analysis based on biorthogonal wavelet transform and a edge detection algorithm based on the fusion technology of wavelet transform and morphological edge detect