Bart Gerard Bernard Barenbrug

Multistep joint bilateral depth upsampling

Depth maps are used in many applications, e.g. 3D television, stereo matching, segmentation, etc. Often, depth maps are available at a lower resolution compared to the corresponding image data. For these applications, depth maps must be upsampled to the image resolution. Recently, joint bilateral filters are proposed to upsample depth maps in a single step. In this solution, a high-resolution output depth is computed as a weighted average of surrounding low-resolution depth values, where the weight calculation depends on spatial distance function and intensity range function on the related image data. Compared to that, we present two novel ideas. Firstly, we apply anti-alias prefiltering on the high-resolution image to derive an image at the same low resolution as the input depth map. The upsample filter uses samples from both the high-resolution and the low-resolution images in the range term of the bilateral filter. Secondly, we propose to perform the upsampling in multiple stages, refining the resolution by a factor of 2×2 at each stage. We show experimental results on the consequences of the aliasing issue, and we apply our method to two use cases: a high quality ground-truth depth map and a real-time generated depth map of lower quality. For the first use case a relatively small filter footprint is applied; the second use case benefits from a substantially larger footprint. These experiments show that the dual image resolution range function alleviates the aliasing artifacts and therefore improves the temporal stability of the output depth map. On both use cases, we achieved comparable or better image quality with respect to upsampling with the joint bilateral filter in a single step. On the former use case, we feature a reduction of a factor of 5 in computational cost, whereas on the latter use case, the cost saving is a factor of 50.

Coherent spatial and temporal occlusion generation

A vastly growing number of productions from the entertainment industry are aiming at 3D movie theatres. These productions use a two-view format, primarily intended for eye-wear assisted viewing in a well defined environment. To get this 3D content into the home environment, where a large variety of 3D viewing conditions exists (e.g different display sizes, display types, viewing distance), we need a flexible 3D format that can adjust the depth effect. Such a format is the image plus depth format in which a video frame is enriched with depth information of all pixels in the video. This format can be extended with an additional layer for occluded video and associated depth, that contains what is behind objects in the video. To produce 3D content in this extended format, one has to deduce what is behind objects. There are various axes along which this occluded data can be obtained. This paper presents a method to automatically detect and fill the occluded areas exploiting the temporal axis. To get visually pleasing results, it is of utmost importance to make the inpainting globally consistent. To do so, we start by analyzing data along the temporal axis and compute a confidence for each pixel. Then pixels from the future and the past that are not visible in the current frame are weighted and accumulated based on computed confidences. These results are then fed to a generic multi-source framework that computes the occlusion layer based on the available confidences and occlusion data.

Enabling Introduction of Stereoscopic (3D) Video: Formats and Compression Standards

After the introduction of HDTV, the next expected milestone is stereoscopic (3D) TV. This paper gives a summary of the new MPEG-C part 3 standard, capable of compressing the 2D+Z format, and shows how it can be used to serve the Is generation of 3DTVs. Furthermore it gives directions on how this standard could be extended to serve also the generations beyond.

Algorithms for division free perspective correct rendering

Well known implemetations for perspective correct rendering of planar polygons require a division per rendered pixel. Such a division is better to be avoided as it is an expensive operation in terms of silicon gates and clock cycles. In this paper we present a family of efficient midpoint algorithms that can be used to avoid division operators. These algorithms do not require more than a small number of additions per pixel. We show how these can be embedded in scan line algorithms and in algorithms that use mipmaps. Experiments with software implementations show that the division free algorithms are a factor of two faster, provided that the polygons are not too small. These algorithms are however most profitable when realised in hardware.

Robust image, depth, and occlusion generation from uncalibrated stereo

Philips is developing a product line of multi-view auto-stereoscopic 3D displays.1 For interfacing, the image-plus-depth format is used.2, 3 Being independent of specific display properties, such as number of views, view mapping on pixel grid, etc., this interface format allows optimal multi-view visualisation of content from many different sources, while maintaining interoperability between display types. A vastly growing number of productions from the entertainment industry are aiming at 3D movie theatres. These productions use a two view format, primarily intended for eye-wear assisted viewing. It has been shown4 how to convert these sequences into the image-plus-depth format. This results in a single layer depth profile, lacking information about areas that are occluded and can be revealed by the stereoscopic parallax. Recently, it has been shown how to compute for intermediate views for a stereo pair.4, 5 Unfortunately, these approaches are not compatible to the image-plus-depth format, which might hamper the applicability for broadcast 3D television.3

Declipse 2: multi-layer image and depth with transparency made practical

To be able to accommodate different viewing conditions (e.g. different display sizes, display types, viewing distance), a 3D format is required where the depth effect can be adjusted. Within MPEG, the image-and-depth format has been standardized. This format is bandwidth efficient, highly flexible and therefore well suited to introduce 3D content into for example the home environment. Extensions to this format have been proposed to enable occlusion handling, by introducing more image-and-depth layers, containing image data located behind the foreground objects, to enable looking around such foreground objects. In the presence of such a multi-layer representation, a next extension is to add transparency information to the layers, allowing for alpha matting (offering significant gains in rendering quality) and inclusion of semi-transparent objects in the scene. Due to the multi-layer nature, it is then still possible to tune the amount of depth. In this paper, we report on our design choices to arrive at the Declipse 2 format, by going through the whole video chain, from content creation, to looking at the effects of video compression and finally real-time 3D rendering and display of such multi-layer content with transparency information, showing the feasibility of the approach under practical conditions.

Fully-Automatic Correction of the Erroneous Border Areas of an Aneurysm

Volume representations of blood vessels acquired by 3D rotational angiography are very suitable for diagnosing an aneurysm. We presented a fully-automatic aneurysm labelling method in a previous paper. In some cases, a portion of a “normal” vessel part connected to the aneurysm is incorrectly labelled as aneurysm. We developed a method to detect and correct these erroneous border areas. Application of this method gives better estimates for the aneurysm volumes.

Shifting of the Aneurysm Necks for Enhanced Aneurysm Labelling

Volume representations of blood vessels acquired by 3D rotational angiography are very suitable for diagnosing an aneurysm. We presented a fully-automatic aneurysm labelling method in a previous paper. In some cases, a portion of a “normal” vessel part connected to the aneurysm is incorrectly labelled as aneurysm. We developed a method to shift the erroneous borders as close to the real aneurysm as possible. Application of this method gives better estimates for the aneurysm volumes.