基于二维Gabor小波和孪生支持向量机的图像识别算法

Image recognition algorithm based on two-dimensional Gabor wavelet and twin support vector machine

为了提高图像识别性能,采用孪生支持向量机用于图像分类识别,并结合二维Gabor小波对图像纹理特征进行提取,借助局部线性嵌入(Locally linear embedding,LLE)降维,以进一步提高图像识别准确率和识别效率。采用二维Gabor小波对图像数据进行有效滤波,获得图像关键纹理特征,然后对大量纹理特征进行LLE降维,以降低维度过高带来的运算量巨大问题,采用孪生支持向量机(Twin support vector machine,TWSVM)对关键纹理特征进行分类,获得图像分类结果并完成图像识别。实验证明,Gabor+LLE+TWSVM方法对图像识别的适用度高,相比常用图像识别算法,通过合理设置二维Gabor小波的尺度和方向参数,并借助LLE有效降维,运用孪生支持向量机可以获得更高的图像识别准确率。

In order to improve the performance of image recognition,twin support vector machine is used to classify and identify images,and the texture features of images are extracted by combining two-dimensional Gabor wavelet.The accuracy and efficiency of image recognition can be further improved by using locally linear embedding(LLE)to reduce dimension.Firstly,the image data are filtered effectively by two-dimensional Gabor wavelet,and then the key texture features of the image are obtained.Then,LLE dimension reduction is carried out for a large number of texture features to reduce the huge computation problem caused by the too high dimension.Finally,the key texture features are classified by twin support vector machine(TWSVM),and the image classification results are obtained and image recognition is completed.The experiment results show that Gabor+LLE+TWSVM has a high applicability to image recognition.Compared with the common image recognition algorithm,the higher accuracy of image recognition can be obtained by using twin support vector machine by setting the scale and direction parameters of two-dimensional Gabor wavelet reasonably and reducing dimension effectively with LLE.