Multiresolution sampling procedure for analysis and synthesis of texture images

By Jeremy S. De Bonet. 1997

This paper outlines a technique for treating input texture images as probability density estimators from which new textures, with similar appearance and structural properties, can be sampled. In a two-phase process, the input texture is first analyzed by measuring the joint occurrence of texture discrimination features at multiple resolutions. In the second phase, a new texture is synthesized by sampling successive spatial frequency bands from the input texture, conditioned on the similar joint occurrence of features at lower spatial frequencies. Textures synthesized with this method more successfully capture the characteristics of input textures than do previous techniques.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Deep Generative Image Models using a Laplacian Pyramid of Adversarial Networks (2015, methodology): “ture synthesis [3, 33, 22], with Portilla & Simoncelli [22] making use of a steerable pyramid wavelet representation [27], similar to our use of a Laplacian pyramid.”
The Synthesizability of Texture Examples (2014, methodology): “color histograms [13], multi-band spatial frequencies [6], and wavelet features [33].”
Simultaneous cartoon and texture image inpainting using morphological component analysis (MCA) (2005, background): “other hand, local statistical analysis and prediction have been shown to perform well at filling in content [3,13,29].”
Hallucinating face by eigentransformation (2005, background): “Some other approaches [5], [6], [8]–[10], [12], [14], [15] are based on learning from the training set containing high- and low-resolution image pairs, with the assumption that high-resolution images are Markov random fields”
Lapped textures (2000, methodology): “However, these methods cannot be readily extended to cover surfaces of arbitrary topology because such surfaces lack continuous parametrizations over the plane.”
Learning how to inpaint from global image statistics (2003, background): “The region can be £lled-in by a texture synthesis engine, e.g. [14] [6] [21] [7] [3].”
Face Hallucination: Theory and Practice (2007, methodology): “(5) contains the prior distribution of a face image p(IH ).”
Texture synthesis by non-parametric sampling (1999, methodology): “De Bonet [1] also uses a multi-resolution filter-based approach in which a texture patch at a finer scale is conditioned on its “parents” at the coarser scales.”
Synthesizing natural textures (2001, background): “Several general texture synthesis algorithms have been developed [5, 7, 9, 13] but until recently both running time [7] and, in many cases, quality [5, 9] of synthesized textures has left much to be desired.”
Multiresolution Markov models for signal and image processing (2002, methodology): “Another very interesting approach to MR modeling for image processing is that developed in [91, 90, 92].”
On seeing stuff: the perception of materials by humans and machines (2001, background): “Heeger and Bergen[11] found that two textures that matched both in terms of pixel statistics and wavelet coefficient statistics tended to look similar, and others have built on this notion to develop successful texture synthesis techniques[12,13].”
TEXTure: Text-Guided Texturing of 3D Shapes (2023, background): “Early works [11, 12, 18, 54] focus on 2D texture synthesis through probabilistic models, while more recent works [13,45,50,53] take a datadriven approach to generate textures using deep neural networks.”
Image Completion Using Efficient Belief Propagation Via Priority Scheduling and Dynamic Pruning (2007, background): “All exemplar-based techniques for texture synthesis that have appeared until now, were either pixel-based [13], [14] or patch-based [15]–[17], meaning that the final texture was synthesized one pixel, or one patch at a time (by simply copying pixels or patches from the observed image, respectively).”
Image quilting for texture synthesis and transfer (2001, background): “De Bonet [3] scrambles the input in a coarse-to-fine fashion, preserving the conditional distribution of filter outputs over multiple scales (jets).”