A robust hybrid image-based modeling system

摘要: This paper presents a new robust image-based modeling system for creating high-quality 3D models of complex objects from sequence unconstrained photographs. The images can be acquired by video camera or hand-held digital without the need calibration. In contrast to previous methods, we integrate correspondence-based and silhouette-based approaches, which significantly enhances reconstruction with few visual features (e.g., uni-colored objects) improves surface smoothness. Our solution uses mesh segmentation charting approach in order create low-distortion parameterization suitable arbitrary genus. A texture is produced first parameterizing reconstructed using approach, projecting sections input onto model, combining them graph-cut technique. Holes due patches are filled novel exemplar-based inpainting method exploits appearance space attributes improve patch search, blends Poisson-guided interpolation. We analyzed effect different algorithm parameters, compared our laser scanning-based existing commercial systems. results indicate that robust, superior other techniques, achieve quality visually not discernible scanner.