Multiscale fusion and aggregation PCNN for 3D shape recovery

Shape from focus (SFF) is a widely used method for recovering the three-dimensional (3D) shape of an object from an image sequence with various focus measure operators. However, most previous studies have focused on evaluating the depth map using a specified focus measure operator based on a single perspective. These methods severely limit the accuracy of the reconstruction results for a complex real scene. In this study, a novel SFF method based on a multiscale fusion perspective is proposed. First, the feasibility of obtaining a mapping relationship between high-frequency coefficients and the depth maps by the stationary wavelet transform (SWT) is discussed. Next, level reduction is introduced to approximate the target depth map using multilevel high-frequency coefficients. Then, an aggregation pulse coupled neural network (a-PCNN) model with variable-size cross sum modified Laplacian (CSML) operators is used as the mapping functions from selected high-frequency coefficients to various window size depth maps. Finally, a hierarchical screening method (HSM) is proposed to yield a more accurate reconstruction result by fusing depth maps with various window sizes. The experimental results demonstrate that the proposed method realizes a more accurate depth map estimation and better surface consistency of the reconstruction results than the compared SFF methods and several advanced multifocus image fusion methods. (C) 2020 Elsevier Inc. All rights reserved.