550 likes | 874 Views
Near-Regular Texture Analysis and Manipulation. Written by: Yanxi Liu Wen-Chieh Lin James Hays. Presented by: Alex Hadas. What we will see today?. Regular, Near-Regular Texture Definition Previous Approaches Near-Regular Texture Analysis Regularity Measurements
E N D
Near-RegularTextureAnalysisandManipulation Written by: YanxiLiu Wen-ChiehLin JamesHays Presented by: • Alex Hadas
What we will see today? • Regular, Near-Regular Texture Definition • Previous Approaches • Near-Regular Texture Analysis • Regularity Measurements • Near-Regular Texture Manipulation • Near-Regular Texture Synthesis Algorithm
Example A: Cloth, Tahiti Example taken from Wikipedia Regular, Near-Regular Texture Definition • Regular Texture – wallpaper-like, congruent 2D tiling whose structural regularity can be completely characterized by 17 wallpaper groups
T2 Example C: Painted porcelain, China T1 Example taken from Wikipedia Regular, Near-Regular Texture Definition • Underlying lattice structure can be represented and generated by a pair of linear independent translations
Example B: Ornamental painting, Nineveh, Assyria Example taken From Wikipedia Regular, Near-Regular Texture Definition • The smallest bounded region that produces (under translation subgroup) simultaneously a covering (no gaps) and a packing (no overlaps) of the texture pattern on 2D plane is called a tile.
Regular, Near-Regular Texture Definition To Algorithm
Regular, Near-Regular Texture Definition • Near-Regular Texture is statistical distortion of a regular, wallpaper like congruent tiling, possibly with individual variations in tile shape, size, color and lighting
Regular, Near-Regular Texture Definition • A Near – Regular Texture p = d(pr), where • pr is regular texture, • d = dgeo×dlight×dcolor, where • dgeo – Geometric Transformation • dlight– Lighting Changes • dcolor – Color Alterations
Brick wall Regular, Near-Regular Texture Definition Examples of Near-Regular Textures Snake Cloth
Regular, Near-Regular Texture Definition Categorization of Near – Regular Textures (NRT)
Previous Approaches • Generative model approach • Cost of model-specific parameter tuning
Previous Approaches • Sample based approach • Neighborhood-based statistical analysis • Non-parametric estimation • Tiling based approach • Only Type I (Lui [2004b] • Only local boundaries preserved, but global near-regularity not addressed (Cohen et al.[2003]
Previous Approaches • Producing regular patterns with translational symmetry by generating tiling boundaries from closed planar contour Escherization [2000] Synthesized results Input Kwatra et al. 2003 Type I
Previous Approaches • Texture transfer problem • Image Analogies [Hertzmann et al. 2001] • Texture Quilting [Efros and Freeman 2001] Synthesized results Input Type II Efros and Freeman 2001
Previous Approaches • Texture replacement on plane • Surface is planar, texture is of type I (Tsin et al. [2001] • Separation illuminance and texture using a non-linear filtering technique (Oh et al[2001]
Near-Regular Texture Analysis • Geometric Deformation Field • Lighting Deformation Field • Color Deformation Field • A Pair of Regularity Measurements
Geometric Deformation Field computer builds 2D lattice User adjusts misplaced points Computer finds optimized lattice Using MFFD for capturing 1 to 1 warping field Represent warping field in HSV space
t1 t1 t1 t1+t2 t1-t2 t2 t2 t2 Geometric Deformation Field
NRT Analysis: Geometric Deformation Field • Represent warping field in HSV space dx dy Color scheme used Displacement Map
Lighting Deformation Field • Straighten the NRT lattice using dgeo • Apply Tsin et al.[2001]’s algorithm for lighting map extraction in the plain • Apply inverse geometric field
Color Deformation Field • PCA method: create set of basis and coefficients
Regularity Measurements • Geometric Regularity • Appearance Regularity
Near-Regular Texture Manipulation • Geometry Deformation Field Manipulation • Texture Replacement • Deformation Field Analogy • Texture Regularity Manipulation
Deformation Field Analogy Geometric Deformation Field Lighting Deformation Field A A’ : Extracted from Input Texture Extracted from Input Texture B B’ : Synthesized from A Result of Deformation Field Analogy
NRT Synthesis Algorithm • Type I NRT only • What is Tile?
NRT Synthesis Algorithm • Minimum tiles set {t i} • Maximum tiles set {Ti} • Centered on half way shifted lattice points
NRT Synthesis Algorithm • Stage 1(analysis) • Determine from a given sample pattern • Determine lattice anchor points {t i} (user controlled) • For each t i construct maximum tile sets T (centered on lattice points) and Th (centered half way)
NRT Synthesis Algorithm • Stage 2 (synthesis) • Start from top left corner with random tile chosen from T • Add tile to the synthesized texture in a scan line along with step When we reach right boundary place tile in direction with step from left most tile in a row
NRT Synthesis Algorithm • Stage 2 (synthesis) (cont.) • At each lattice or half-way lattice point select T or Th tile set and pick one of the best tiles. Error function value is less that threshold
NRT Synthesis Algorithm Error Function Distance Function Red values of the pixel Blue values of the pixel Green values of the pixel
NRT Synthesis Algorithm • Stage 2 (synthesis) (cont.) • Register selected candidate tile using a correlation-based method • Use dynamic programming to “stitch” the overlapping tiles. Apply it separately to horizontal and vertical directions
NRT Synthesis Algorithm • Stage 2 (synthesis) (cont.) • When pasting a tile to existing image apply blending where dynamic programming may have conflicting decisions. • Repeat steps 2-6 until the whole image is synthesized
selected tile synthesized tile depends on distance of pixel to the boundary NRT Synthesis Algorithm