Research Of Depth Restoration Algorithm For Core Defocused Image

In the field of computer vision,three-dimensional reconstruction is a highly researched area,and obtaining depth information of objects in the scene is crucial for three-dimensional reconstruction.Therefore,acquiring depth information has always been a significant aspect in the field of three-dimensional reconstruction.Recovering the three-dimensional information of rock cores is of great importance for petroleum exploration and geological research.It can be used for sedimentary facies analysis,studying the geological environment during formation,as well as for specialized research such as lithological characterization,stratigraphic age determination,and stratigraphic correlation.However,during the processes of drilling,transportation,processing,and storage,rock cores can deviate from their ideal cylindrical shape due to improper handling,resulting in various sizes of defects and significant unevenness in their upper and lower cross-sections.Recovering the three-dimensional structure of such irregular rock cores is relatively challenging,making data acquisition difficult and affecting the quality and accuracy of depth recovery.Additionally,the imaging devices used for rock core capture employ high-resolution fixed-focus lenses,which have limited depth of field,leading to inevitable defocused regions in the captured images.Lastly,there is an urgent need to develop a fast depth recovery method based on the characteristics of existing images,as there is a large amount of scanned images of rock core side views and corresponding upper and lower crosssectional scans available.There are various methods available for depth recovery of rock cores,and the effectiveness of these methods varies.For instance,laser ranging in active depth recovery methods can be used to scan rock core samples and obtain corresponding depth information with high accuracy.However,this method has higher environmental requirements and typically requires measurement on smooth and uniform surfaces.Moreover,laser scanners are expensive,and the measurement process is slow,consuming a significant amount of time.Therefore,passive depth recovery methods are more commonly used and practical in real-world applications.Based on the characteristics of existing rock core samples and the limitations of current image capture devices,this study compares several passive depth recovery algorithms for rock core depth recovery,highlighting their advantages and disadvantages.Eventually,a passive depth recovery method based on Defocus Image Depth(DFD)algorithm is chosen,in combination with a deep neural network model,to obtain relative depth variations in rock core images.The specific approach involves capturing a set of defocused images of the rock core at equidistant distances from the camera.The rock core’s all-in-focus image is obtained through steps such as focus stack alignment,focus measurement,and all-in-focus image stitching.Subsequently,the all-in-focus image is gradually blurred using point spread parameters that increase incrementally at equidistant intervals.20x20-pixel image patches,sliced from the blurred images,are then fed into a residual regression neural network for training.The trained neural network model can be used to recover the depth of a single untrained defocused image of a rock core.Through quantitative and qualitative comparisons,the superiority of this approach is demonstrated.Compared to existing networks such as D3 net,which achieve good depth recovery results,the experiments conducted on multiple sets of rock core images show that the proposed method provides better depth recovery results.The method accurately recovers depth information from rock core images in the experimental environment,with more prominent details.