Depth-of-field rendering with multiview synthesis

Abstract

Figure 1: Our method achieves DOF blur effects comparable to accurate solutions in real time and avoids postprocessing artifacts. AbstractWe present a GPU-based real-time rendering method that simulates high-quality depth-of-field effects, similar in quality to multiview accumulation methods. Most real-time approaches have difficulties to obtain good approximations of visibility and view-dependent shading due to the use of a single view image. Our method also avoids the multiple rendering of a scene, but can approximate different views by relying on a layered image-based scene representation. We present several performance and quality improvements, such as early culling, approximate cone tracing, and jittered sampling. Our method achieves artifact-free results for complex scenes and reasonable depth-of-field blur in real time.

Real-time depth of field using multi-layer filtering

“…Regarding image quality, we compare our results to the approach by Lee et al [2009] when run with sufficient lens samples. A comparison for the difficult overlap case is shown in Figure 11 where the Table 1: Detailed times (in ms) for the scene shown in Figure 1 with the text in focus, rendered at 720p.…”

“…Rendering resolution was 720p using one mip-mapping step followed by three stack filtering steps, measured on a GeForce GTX 780. Figure 12: We compare our approach (top, 52.6 frames/sec) to the reference depth of field blur by Lee et al [2009] (center, 256 samples per pixel). The blurred neon text shines through the outof-focus railing.…”

“…Even if the method is optimized for GPUs, no real-time rates are reported for reasonable resolutions. A real-time method that can also generate transparent, nearby objects is the method by Lee et al [2009]. In their approach, lens blur is simulated similarly to offline-rendering by sampling the lens with a number of rays.…”

“…Note that our approach does not suffer from pointsampling artifacts as ray tracing based methods do (e.g. [Lee et al 2009]) as exemplified in Figure 12 (bottom). Our approach also seamlessly integrates semi-transparent surfaces such as the glass elements, which are not easily addressed by other (layered) methods.…”

“…Figure 11: A comparison of our approach (top, 49.8 frames/sec) to the reference depth of field blur according toLee et al [2009] (bottom, 128 samples per pixel, 15.8 frames/sec). Note how fine structures in the front reveal the in-focus text behind.…”