Facade Texture Inpainting

This work presents the method to generate facade texture eliminating the occlusions of foreground objects from multi-view images.

Rectification

Extract rectified facade region via vanishing points.

Input image and R.O.I. selection

Line segments for vanishing points extraction

Rectified facade texture

Repetitive pattern analysis

  • Analyze the repetitive pattern for given facades.
  • Analyze the structural information around of the occluded region.

Feature extraction : From the left, input, GFTT (or FAST) features, horizontal matches and vertical matches.

Region growing/merge - Horizontal : From the left, iteration 10, 50, 100 and final results.

Region growing/merge - Vertical : From the left, iteration 10, 50, 100 and final results.

Repetitive patterns

Occlusion Detection

Detection occluded regions of each facade from given images, the camera information of each image and 3D building model.

  • Graph-cut based segmentation : From the left, rectified image, user stroke, over-segmentation and occlusion mask.
  • Flood-fill based segmentation : From the left, rectified image, user stroke and occlusion mask.

Multi-view integration

Generate the fronto-parallel texture via integration of the results of the multi-view.

Publication

  • Seon Ho Oh and Soon Ki Jung, Facade Texture Inpainting for 3D City Reconstruction, Korea Computer Congress 2012, 2012.