Quality Improvement for Intermediate Views Using Example-Based Super-Resolution

We propose an algorithm that improves the quality of intermediate-view images obtained by mapping the average color of multiple-view images onto a visual hull. The algorithm estimates a real image from an intermediate-view image using example-based super-resolution. Estimation is performed using relationships between two kinds of training images: an ideal image whose quality is as good as a real image and an intermediate view image viewed from the same viewpoint. High frequencies of the real image of an intermediate-view are estimated using the relationship between high frequencies of an ideal image and those of a corresponding intermediate-view image. Experimental results show that the quality of the resulting images is almost as good as that of ideal images. Quality Improvement for Intermediate Views Using Example-Based Super-Resolution Polygon Reconstructed model