论文部分内容阅读
针对基于单一特征进行可通行性地形分类效果差的问题,提出了一种融合多可视化特征的地形分类算法.首先通过实验选出了分类效果较好的YIQ颜色空间并在此空间提取颜色特征,然后引入一种新的能量定义方法对离散余弦变换(DCT)纹理特征提取法加以改进,由实验得出改进的DCT纹理特征及小波(Coiflets-4)纹理特征可取得较好的分类效果.将上面3种特征加以融合并用主成分分析法(PCA)进行降维处理,利用高斯混合模型(GMM)作为分类器,在由VisTex标准数据库所生成的马赛克图像和真实的野外环境图像中进行实验,结果令人满意.
Aiming at the problem of poor terrain classification based on single feature, a terrain classification algorithm based on multi-visualization features is proposed.Firstly, the YIQ color space with good classification performance is selected by experiments and the color features are extracted in this space, Then, a new energy definition method is introduced to improve the DCT texture feature extraction method, and the improved DCT texture feature and the wavelet texture feature (Coiflets-4) can be obtained by experiments. The above three features are fused and reduced by principal component analysis (PCA). Using Gaussian Mixture Model (GMM) as a classifier, experiments are carried out on the mosaic images generated by the VisTex standard database and the real field environment images. The result is satisfactory.