基于Shearlet变换的井下图像差异性特征提取方法

Otherness feature extraction method for underground image based on Shearlet transform

针对井下收集的人脸图像易受煤尘干扰且一般特征提取方法对噪声较敏感的问题,提出一种基于Shearlet变换的井下图像差异性特征提取方法。首先利用Shearlet变换将图像进行多尺度多方向分解,然后对同一尺度的各方向子图利用实部特征进行编码融合,进而根据各尺度子图的Shannon熵值赋予不同权值进行再融合,最后对低频子图和融合后的高频子图利用Shearlet逆变换重构得到差异性图像。实验结果表明,该方法具有较好的客观评价指标与主观效果。

For the problem that face images collected underground are susceptible to dust interference and most feature extraction methods are sensitive to noise, an otherness feature extraction method for underground image based on Shearlet transform was proposed. First, Shearlet transform was used for multi-directional and multi-scale image decomposition; then each directional sub-graphs with the same scale were encoded and fused, and furthermore fusion was conducted by giving different weights according to Shannon entropy of the subgraphs; finally, Shearlet inverse transform was used to reconstruct otherness image of low-frequency sub-graph and fused high-frequency sub-graph. The experimental results show that the method has a good objective and subjective evaluation.