摘要
对网络中三维景观图像样品进行查询,在提高图像利用率方面具有重要作用。由于网络的开放性及可扩展性,使得三维景观图像复杂且质量易变;传统查询方法主要通过三维景观图像视觉信息进行查询,忽略了图像转载过程中产生的尺度变化和噪声的干扰,导致查询不完整,制约了图像的使用率,提出基于极端学习机的三维景观图像样品查询方法。采用非采样轮廓波变换对图像不同尺度、同方向的高频方向子带求多尺度积,考虑到三维景观图像旋转后各个方向子带在图像中的能量所占比不会发生变化,将三维景观图像各方向子带的能量占比作为概率矢量,各方向子带的粗糙度作为权值求取三维景观图像的加权信息熵,提取三维景观图像的颜色和形状特征,在三维景观图像边缘信息特征查询反馈阶段,将三维景观图像最大边际准则引入到极端学习中,通过分析极端学习机的类内离散度和类间离散度获得包含图像判别边缘信息的分类模型,以提高三维景观图像样品查询能力。实验结果表明,该方法对旋转变换适应性较强,显著提高了三维景观图像样品查询效率。
A query method of three - dimensional landscape image samples based on extreme learning machine is proposed. The non - sampling contourlet transform image is used to seek the multi - scale product of high frequency directional sub band of image with different scales and in the same direction with high frequency directional sub for multi - scale product. Considering that the energy ratio of different sub - bands in the image after three - dimensional landscape image rotates would not change, the energy ratio of sub - band in three - dimensional landscape image on each direction is taken as the probability vector, and the roughness of sub bands in each direction is taken as weigh- ted information entropy of three - dimensional landscape image. The color and shape features of three - dimensional landscape image are extracted. In the stage of query feedback of three - dimensional landscape image edge features, the equimarginal principle of three - dimensional landscape image is introduced to the extreme learning. Through the analysis of within - class scatter and between - class scatter of extreme learning machine, the classification model including the edge information of image discriminant is obtained, thus the query capability of three - dimensional landscape image sample is improved. Simulation result shows that the proposed method has strong adaptability on the rotation transform, which improves the query efficiency of samples of three - dimensional image.
出处
《计算机仿真》
北大核心
2017年第12期313-317,共5页
Computer Simulation
关键词
网络
三维景观图像
样品
边缘信息
查询
Network
Three - dimensional landscape image
Sample
Edge information
Query
