基于图像特征提取的算法设计与应用

【作者】 张鹏；

【导师】 朱小芳； 杜新立；

【作者基本信息】 电子科技大学 ， 软件工程， 2011， 硕士

【摘要】 随着工业的集约化发展,表面产品的生产量持续大幅度增长。在生产中,对表面加工质量高水平的苛求,尤其是一致性的要求,对图像识别技术提出了更高的要求。本文研究了表面图像预处理、分类识别技术、特征提取、模式识别问题,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。本文最后采用“车牌字符识别实验”对图像分类识别进行实现。本文从颜色特征角度来描述,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。实验结果证明:根据图像的纹理特征和颜色特征,运用数字图像处理技术,来解决图像的分割和识别等问题,是行之有效的途径。更多还原

【Abstract】 Along with industrial intensive development, surface production continued to grow substantially. In production, to quality of surface processing of high levels of demanding, especially the consistency requirements, on the image recognition technology has put forward higher requirements.This paper studies the surface image preprocessing, feature extraction, classification and recognition technology, pattern recognition problem, traditional filtering algorithm in the suppression of noise at the same time, also on the image edges and details are relatively large damage, so that the edge of image and blur detail problem, put forward the weighted directional smoothing algorithm. And in the image segmentation on the fusion of several segmentation methods, put forward a kind of improvement based on biorthogonal wavelet transform multiresolution image fusion method based on fusion and wavelet transform and morphological edge detection algorithm, optimization of the segmentation effect, for the follow-up feature extraction is to lay a good foundation.Finally, using the" license plate character recognition experiment" to image classification to achieve. This article from the color features are described in terms of. According to the parameters of the distribution, selection of less standard deviation parameters as classifier input feature vector; and using the principal component analysis method for feature extraction, texture feature dimension reduction, elimination patterns of correlation among features, highlighting their differences, meet the needs of the recognition layer.The experimental results show that: according to the image texture features and color features, using digital image processing technology, to solve the problem of image segmentation and recognition, is effective way.更多还原