融合多可视化特征的可通行性地形分类

Traversability classification based on multi-visual features fusion

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.

A terrain classification method based on multi-visual features fusion was proposed,which improves the poor performance of classification method based on single feature.First,YIQ color feature which can get better classification performance in the experiment was chosen to compute the color feature.Second,texture feature extraction method based on discrete cosine transform(DCT) was improved by introducing a new energy define method.In the experiment texture features based on the improved discrete cosine trans...