基于空域与频域结合的聚焦形貌恢复

Shape from focus based on combination of spatial and frequency domains

  • 摘要: 针对单一聚焦评价函数存在重建精度低,评价曲线分辨力差的缺点,提出一种空域与频域结合的聚焦形貌恢复方法。首先,使用空域拉普拉斯算子和频域非下采样小波变换对图像进行聚焦评价,将非下采样小波变换聚焦评价值作为权重因子,对拉普拉斯算子评价值进行修正;然后使用修正的聚焦评价值重构表面形貌,获得初始深度图;接着计算出基于拉普拉斯算子的深度图,并将该深度图作为引导图对初始深度图进行引导滤波,获得最终的深度图。使用仿真与实验验证,结果表明:该方法在仿真与实验中均有较好的效果,评价曲线分辨力明显提高,对仿真模型的重建有更低的均方根误差和更高的相关系数,在实验中三维重建的模型与实物误差更小,可有效提高重建精度。

     

    Abstract: The single focus evaluation function has the disadvantages of poor resolution and low reconstruction accuracy. To solve this problem, a focusing topography restoration method combining spatial domain and frequency domain was proposed. Firstly, the spatial Laplacian operator and frequency domain non-subsampled wavelet transform were used to evaluate the image focus, and the non-subsampled wavelet transform focus evaluation value was used as the weight factor to correct the Laplacian evaluation value. The initial depth map was obtained by using modified focused evaluation to reconstruct the surface topography. Then the depth map based on Laplacian operator was calculated, and the depth map was used as the guide map to guide the initial depth map and filter the final depth map. Simulation and experimental verification show that the proposed method has good results in both simulation and experiment. The resolution of evaluation curve is obviously improved, and the reconstruction of simulation model has lower root-mean-square error and higher correlation coefficient. In the experiment, the error between the 3D reconstructed model and the real object is smaller, which can effectively improve the accuracy.

     

/

返回文章
返回