高斯颜色模型在瓷片图像分类中的应用

Porcelain shard images classification based on Gaussian color model

由于RGB颜色空间不能很好贴近人的视觉感知,同时也缺少对空间结构的描述,因此采用兼顾颜色信息和空间信息的高斯颜色模型以获取更全面的特征,提出了一种基于高斯颜色模型和多尺度滤波器组的彩色纹理图像分类法,用于瓷器碎片图像的分类。首先将原始图像的RGB颜色空间转换到高斯颜色模型;再用正规化多尺度LM滤波器组对高斯颜色模型的3个通道构造滤波图像,并借助主成分分析寻找主特征图,接着选取各通道的最大高斯拉普拉斯和最大高斯响应图像,与特征图联合构成特征图像组用以进行参数提取;最后以支持向量机作为分类器进行学习和分类。实验结果表明,与基于灰度的、基于RGB模型的和基于RGB_bior 4.4小波的方法相比,本文方法具有更好的分类结果,其中在Outex纹理图像库上获得的分类准确率为96.7%,在瓷片图像集上获得的分类准确率为94.2%。此方法可推广应用到其他彩色纹理分类任务。

Since the RGB color space does not closely match the human visual perception and has no ability to describe the spatial structures, the Ganssian celor model, which uses the spatial and color information in an integrated model, is used to obtain more complete image features. A color-texture approach based on the Gaussian color model and a multi-scale filter bank is introduced to classify the porcelain shard images. First, the RGB color space of the image is transformed into the Gaussian color model and then the normalized multi-scale LM filter bank is used to construct the filtered images on three channels. Afterwards, the primary feature images are found by using principal components analysis mid the maximum re- sponses of the Laplacian of Ganssian filters and Gaussian filters are separately selected. These images compose a feature im-age set, in which the feature parameters are extracted. Finally, a support vector machine is used to learning and classifica-tion. From experimental results, the proposed method is better than gray-based method, RGB-based method and RGB_ bior 4. 4 wavelet based method. It can achieve a classification accuracy of 96. 7% on Outex texture database and a classifi- cation accuracy of 94. 2% on porcelain shard images. This method can be used in other color texture classification tasks.