Improved edge detection for EWOC depth upscaling

The need for accurate depth information in three-dimensional television (3DTV) encourages the use of range sensors, i.e. time-of-flight (ToF) cameras. Since these sensors provide only limited spatial resolution compared to modern high resolution image sensors, upscaling methods are much needed. Typical depth upscaling algorithms fuse low resolution depth information with appropriate high resolution texture frames, taking advantage of the additional texture information in the upscaling process. We recently introduced a promising upscaling method, utilizing edge information from the texture frame to upscale low resolution depth maps. This paper examines how a more thorough edge detection can be achieved by investigating different edge detection sources, such as intensity, color spaces and difference signals. Our findings show that a combination of sources based on the perceptual qualities of the human visual system (HVS) leads to slightly improved results. On the other hand these improvements imply a more complex edge detection.