This paper introduces a joint bilateral upsampling (JBU) method for image analysis and enhancement tasks such as tone mapping, colorization, stereo depth, and photomontage. These tasks often require computing a solution over the pixel grid, but computational and memory costs make it necessary to compute a solution on a downsampled image and then upscale it. Traditional upsampling methods assume a smoothness prior for interpolation, but JBU leverages the high-resolution input image as a prior to produce better results. The JBU method applies a bilateral filter, which preserves edges while smoothing, to upsample the low-resolution solution. It uses both a spatial filter and a range filter, with the range filter applied to the high-resolution input image. This allows the method to take advantage of the high-resolution image's information to produce more accurate results. The paper demonstrates the effectiveness of JBU on various applications. For tone mapping, JBU produces results that are visually and numerically closer to the ground truth compared to traditional methods. For colorization, JBU avoids color spill over edges. For stereo depth, JBU produces more accurate depth maps. For graph-cut based image operations, JBU improves the quality of label maps. The complexity of JBU is O(Nr²), where N is the output image size and r is the domain filter radius. The performance is proportional to the output size and not to the upsampling factor, as the domain filter is applied to the low-resolution solution. The method is efficient and can handle large images, as demonstrated by its application to multi-gigapixel images. The paper concludes that JBU is a promising technique for image analysis and enhancement tasks, and that it can be applied to other domains beyond image processing. The method is local, has a small memory footprint, and can be applied to large images with efficient memory management.This paper introduces a joint bilateral upsampling (JBU) method for image analysis and enhancement tasks such as tone mapping, colorization, stereo depth, and photomontage. These tasks often require computing a solution over the pixel grid, but computational and memory costs make it necessary to compute a solution on a downsampled image and then upscale it. Traditional upsampling methods assume a smoothness prior for interpolation, but JBU leverages the high-resolution input image as a prior to produce better results. The JBU method applies a bilateral filter, which preserves edges while smoothing, to upsample the low-resolution solution. It uses both a spatial filter and a range filter, with the range filter applied to the high-resolution input image. This allows the method to take advantage of the high-resolution image's information to produce more accurate results. The paper demonstrates the effectiveness of JBU on various applications. For tone mapping, JBU produces results that are visually and numerically closer to the ground truth compared to traditional methods. For colorization, JBU avoids color spill over edges. For stereo depth, JBU produces more accurate depth maps. For graph-cut based image operations, JBU improves the quality of label maps. The complexity of JBU is O(Nr²), where N is the output image size and r is the domain filter radius. The performance is proportional to the output size and not to the upsampling factor, as the domain filter is applied to the low-resolution solution. The method is efficient and can handle large images, as demonstrated by its application to multi-gigapixel images. The paper concludes that JBU is a promising technique for image analysis and enhancement tasks, and that it can be applied to other domains beyond image processing. The method is local, has a small memory footprint, and can be applied to large images with efficient memory management.