基于视觉信息的图像特征提取算法研究

Research on the extraction algorithm of image feature based on visual information

自计算机视觉产生开始,视觉信息则自动成为其处理的对象。纹理特征作为视觉信息的重要部分,成为图像特征提取的重点。针对纹理特征提取中,传统LBP算法作为一种基于灰度的算法,仅仅局限对低分辨率图片,其识别率通常比较低的问题,提出一种基于小波变换的LBP算法。该算法利用小波变换的局部特性、方向性和多方向性等特点,并结合LBP算法自身具有的旋转不变形、灰度不变形等特点,从而通过小波变换对高分辨率图片的分解,再通过局部二值模式的优势,从而实现对图像纹理特征的提取。最后通过MATLAB软件编程与ORL数据库的对比测试,改进算法的识别率明显优于传统LBP算法。

The visual information is automatically processed by the visual information from computer vision. As the important part of visual information, texture feature is the key of image feature extraction. For texture feature extraction, the traditional LBP algorithm as a kind of algorithm based on gray, confined to low resolution images and the recognition rate is usually low, put forward a LBP algorithm based on wavelet transform. The algorithm using wavelet transform local characteristics, direction and multi direction of, combined with LBP algorithm has its own rotation deformation, intensity and deformation characteristics, to through wavelet transform on high resolution image decomposition, through the second part value mode advantage, in order to achieve the extraction of image texture features. Through the comparison of MATLAB software programming with ORL database, the recognition rate of the improved algorithm is obviously better than the traditional LBP algorithm.