视差估计下VR图像几何特征数字化提取

Geometric Feature Extraction of VR Image Based on Disparity Estimation

为了快速理解图像信息,提高可视化识别分类效果,提出视差估计下VR图像几何特征数字化提取。将图像几何特征分为面积、周长、质心与延伸方向等类型,利用视差估计法获取图像相邻块间相似尺度,设定阈值,选取最佳参考视点;根据参考视点,引入高斯卷积核确定空间内核,构建尺度空间,保留图像边缘信息;定义候选点,初步划分关键点区间,针对候选点空间函数值,通过阈值比较,过滤出对比度较低的点,建立关键点集合;利用离散Gabor小波变换方法,得出Gabor变换系统数均值和方差,组成几何特征向量;结合最大能量值实现所有特征空间的向量排序,完成几何特征数字化提取。仿真结果表明,上述方法可利用较少的特征点准确提取出几何特征,更有利于图像识别分类。

For quickly grasping the image information and improving the classification effect of visualrecognition, this paper puts forward the digital extraction of geometric features of VR image based on disparity estima-tion. The geometric features of the image were divided into area, perimeter, centroid and extension direction. The dis-parity estimation method was used to obtain the similarity scale between adjacent blocks of the image, thus setting thethreshold and select the best reference view. Gaussian convolution kernel was introduced to determine the spatial ker-nel and construct the scale space via selecting the reference viewpoint, retaining the image edge information. Candi-date points were defined to divide the interval of key points. According to the results of threshold comparison, thepoints with low contrast were filtered out, and the set of key points was also established. Discrete Gabor wavelet trans-form was applied to obtain the mean and variance of the number of Gabor transform systems to form geometric eigen-vectors. The maximum energy value was adopted to realize the vector ordering of all feature spaces. Finally, the ex-traction of geometric features was achieved. The simulation results show that this method is more conducive to imagerecognition and classification, because it can use less feature points to extract geometric features accurately.