GPU-Friendly Laplacian Texture Blending

TL;DR

The paper tackles visible seams and contrast loss in texture blending by introducing a GPU-friendly Laplacian pyramid blending method that operates directly on the existing texture mipmaps with no precomputation. It distributes blending across Laplacian levels using a mask with level-wise Gaussian content, enabling perceptually natural transitions and preservation of fine details. Key contributions include an inline Laplacian construction from mipmaps, a frequency-aware approach to blending radii, a variance-based analysis guiding level choice, and practical GPU optimizations such as level skipping and dynamic mask levels. The method supports material layering and procedural tiling (e.g., hex tiling) with minimal memory overhead, at the cost of additional texture samples and arithmetic, making it suitable for real-time rendering applications with careful profiling.

Abstract

Texture and material blending is one of the leading methods for adding variety to rendered virtual worlds, creating composite materials, and generating procedural content. When done naively, it can introduce either visible seams or contrast loss, leading to an unnatural look not representative of blended textures. Earlier work proposed addressing this problem through careful manual parameter tuning, lengthy per-texture statistics precomputation, look-up tables, or training deep neural networks. In this work, we propose an alternative approach based on insights from image processing and Laplacian pyramid blending. Our approach does not require any precomputation or increased memory usage (other than the presence of a regular, non-Laplacian, texture mipmap chain), does not produce ghosting, preserves sharp local features, and can run in real time on the GPU at the cost of a few additional lower mipmap texture taps.

Figures (12)

• Figure 1: Different blending radii can affect the visual look of blended tiled textures. We can observe either unnatural, harsh transitions (left) or significant blurriness and contrast loss (middle). Furthermore, the rightmost example shows significant ghosting and overlap of distinct details.
• Figure 2: Left: Direct texture blending with a small radius. Middle: Direct texture blending with a large radius. Right: Laplacian texture blending using three Laplacian levels with a wide radius blending low-frequency details and a narrow radius blending high-frequency details.
• Figure 3: Top to bottom: Noise textures with a different frequency content, from low to high frequencies. Left to right: Different blending radii between two textures. Different frequencies of noise require different transition radii for the most natural appearance. A small blending radius produces visible discontinuities on low-frequency content, while a wide radius causes contrast and detail loss on high-frequency textures.
• Figure 4: A toy 1D example demonstrating how, with Laplacian blending, regions of variance loss are distributed between different levels.
• Figure 5: From left to right: Blending with three to seven Laplacian levels. A higher level count increases the smoothness and perceived continuity of the transition, but all preserve sharp contrast and high-frequency features without visible ghosting. The transitions above five Laplacian levels are very smooth, losing some of the distinct identity of the source textures. This can be both an advantage or undesirable from an artistic perspective depending on the use case.