Interactive HDR Image-Based Rendering from Unstructured LDR Photographs

Loubna Lechelek (1) , Daniel Meneveaux (2, 3) , M. Ribardière (2, 3) , Romuald Perrot (2, 3) , Chaouki Babahenini (1)

1 BISKRA - Université Mohamed Khider de Biskra
2 Université de Poitiers - Faculté de Sciences fondamentales et appliquées
3 XLIM-ASALI - Synthèse et analyse d'images

This paper proposes an interactive High Dynamic Range (HDR) image-based rendering system, dedicated to manually acquired photographs. It only relies on a point-based geometric proxy and the original photographs, calibrated using a standard structure-from-motion process. First, a depth map is estimated for each new rendered viewpoint, based on the point cloud. Second, pixel values are reconstructed from the original photographs, using a blending model that also handles occlusion. Our system can be used for producing HDR images from several series of unaligned photographs with different exposures. As shown in the results, it proves efficient with various types of objects, implemented in WebGL, making it practical for many purposes.

Mots clés: Image-based rendering; interactive rendering; Point-based proxy; high-dynamic range images

Domaines: Synthèse d'image et réalité virtuelle [cs.GR]

Loubna Lechelek, Daniel Meneveaux, M. Ribardière, Romuald Perrot, Chaouki Babahenini. Interactive HDR Image-Based Rendering from Unstructured LDR Photographs. Computers and Graphics, 2019, 84, ⟨10.1016/j.cag.2019.07.010⟩. ⟨hal-02278295⟩