多尺度引导滤波及其在去雾中的应用

发布时间：2018-10-17 08:21

【摘要】：将引导滤波与提升小波相结合提出了一种多尺度引导滤波方法,以实现在平滑图像细节的同时保持图像边缘不模糊。该方法通过提升小波法对将图像进行多尺度分解,即将信号分解成一个低频子带和多个高频子带。在提升小波重构过程中,利用引导滤波平滑每个尺度的低频信息并保持其边缘不模糊。最后,针对滤波后残余的细节,对提升小波重构后的平滑图像再次进行引导滤波,以便进一步平滑图像细节。将多尺度引导滤波应用于暗通道去雾先验理论并进行了主、客观评价。结果显示:多尺度引导滤波能够深层次平滑图像细节,保持边缘完整性,从整体上提高了图像的对比对和视觉效果,有效恢复了场景信息并保留场景的边缘信息。另外,该方法改善了客观评价指标,其对比度增强系数指标平均提升了0.1以上,场景结构相似度平均提升了1以上,而LOE(Lightness Order Error)参数降低了10以上,满足了去雾应用的视觉需求。
[Abstract]:A multi-scale guided filtering method is proposed by combining the bootstrapping filter with lifting wavelet in order to smooth the image details while keeping the edge of the image not blur. This method decomposes the image into one low frequency subband and several high frequency subbands by lifting wavelet method. In the process of lifting wavelet reconstruction, the low frequency information of each scale is smoothed by guided filter and its edge is not blurred. Finally, according to the residual details of the filter, the smooth image reconstructed by lifting wavelet is guided again to further smooth the image details. The multiscale guided filter is applied to the priori theory of defogging in dark channels and the subjective and objective evaluation is carried out. The results show that the multi-scale guided filter can smooth the details of the image at a deep level, maintain the edge integrity, improve the contrast and visual effect of the image, restore the scene information and retain the edge information of the scene effectively. In addition, the objective evaluation index is improved, the contrast enhancement coefficient is improved more than 0.1 on average, the scene structure similarity is improved by more than 1 on average, and the LOE (Lightness Order Error) parameter is reduced by more than 10, which meets the visual requirement of defog application.
【作者单位】：中国科学院长春光学精密机械与物理研究所;中国科学院大学;
【基金】：国家自然科学基金资助项目(No.61401425)
【分类号】：TP391.41

【相似文献】