Seam-Boundary Defect Problem

• Seam-boundary defect problem is defined as the interdependence of seam and boundary data that invalidates standard local constructions across various fields.
• It spans diverse applications from image stitching and forensic seam traces to fault-tolerant quantum computation and statistical mechanics.
• Methodologies such as effective data reconstruction and strata compatibility, along with categorical and variational formulations, address these complex interface challenges.

The seam-boundary defect problem denotes a family of technically distinct problems in which a seam, interface, or artificial cut meets a boundary or defect set in a way that invalidates the standard local construction. In image stitching, the seam through an overlap region can generate visible ghosting, color transitions, or structure breakage when local misalignment remains after alignment (Li et al., 2017). In defect-adaptive lattice surgery, the intended seam can intersect deformed boundaries, disabled checks, and gauge-inferred super-stabilizers, so that the textbook product-of-seam-checks observable no longer exists (Min et al., 28 Apr 2026). In logarithmic minimal models, seams fused to a Robin boundary control bulk defect content and determine conformal spectra with half-integer boundary labels (Bourgine et al., 2016). The phrase is therefore best understood as a cross-disciplinary label for problems in which seam data and boundary data cannot be treated independently, and must instead be reconstructed jointly from admissible local structure.

1. Shared structural pattern

Across the cited literature, the seam may be a graph-cut boundary in an image overlap, a merge line in lattice surgery, an $r$-type or $s$-type seam fused to a boundary in a loop model, a domain wall ending on a gapped boundary, an embedded Seifert surface with boundary knot, a line defect in a shell, or an artificial outer boundary in an atomistic simulation. The associated defect may be local misalignment, disabled ancillas, boundary data-qubit defects, conserved loop defects, junction excitations, dislocations, metric anomalies, or compactly supported inhomogeneities. This suggests a common template: a nominally simple seam rule becomes invalid once boundary irregularities destroy support, modify admissible labels, or redistribute the observable over auxiliary data.

Two recurrent mechanisms organize the subject. The first is reconstruction from effective data. In defect-adaptive lattice surgery, the requested parity is synthesized from an admissible effective seam family and pre-merge constraints rather than read off from a uniform strip of seam checks. In topological quantum computation with gapped boundaries and boundary defects, the endpoint of a seam is classified by boundary-defect objects such as $\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$, and fusion proceeds by relative tensor product. In the factorization-algebra formulation, a defect along $D \subset M$ is a factorization algebra $B$ on $M$ together with an isomorphism $B|_{M-D} \cong A$, so the singular locus is encoded by an extension of observables rather than by a naive restriction (Cong et al., 2017, Contreras et al., 2022).

The second is compatibility across strata. In finite-group Dijkgraaf–Witten theory with an embedded Seifert surface and boundary knot, the data are organized by a three-object category with objects $1<2<3$ for line, seam, and bulk, together with a partially degenerate cocycle condition chosen precisely to survive the stratified Pachner moves involving the seam–boundary junction. In integrable defect networks, purely transmitting sewing conditions at a junction are so restrictive that the number of branches meeting at the junction must be even. In each case, the seam-boundary defect problem is not just local singular behavior; it is a constraint problem for the entire stratified configuration (Lee et al., 2015, Corrigan et al., 2020).

2. Image stitching, seam prediction, and forensic seam traces

In consumer-level image stitching, seam-boundary defects arise because global alignment leaves local misalignment, parallax, exposure differences, and geometric distortion in the overlap region. When the final mosaic is composed by selecting a seam through the overlap, the seam boundary can exhibit ghosting, edge discontinuities, color or exposure transitions, and structure breakage. The standard graph-cut formulation minimizes

$$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$

but a linear pixelwise difference cost does not model the nonlinearity and nonuniformity of human perception. The perception-based formulation replaces the linear color difference by a sigmoid metric,

$s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$

and weights pairwise costs by saliency through

$s$0

Here $s$1 is estimated by Otsu’s method on the histogram of $s$2, $s$3, $s$4, and $s$5 is normalized to $s$6. A user study in which 15 participants evaluated 15 groups of stitching results reported that the perception-based approach won the majority of comparisons (Li et al., 2017).

A later formulation recast seam prediction as mask prediction. DSeam takes two warped images $s$7, Sobel edge maps, and an edge-difference image $s$8, and predicts complementary soft masks $s$9 with $\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$0. The seam is the common boundary between the binary masks. Training avoids ground-truth seams by using a Selection Consistency Loss,

$\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$1

with $\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$2, $\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$3, and patch size $\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$4. On the UDIS-D dataset, DSeam achieved lower $\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$5 than dynamic programming and GraphCut across patch sizes $\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$6 from 2 to 15, while running about 15 times faster than OpenCV GraphCut and reaching about $\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$7 FPS at $\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$8 inputs (Cheng et al., 2023).

A distinct forensic use of seam-boundary defects appears in seam-carving detection. Seam removal stitches non-adjacent pixels together, whereas seam insertion duplicates local structure. ILFNet treats the resulting seam-boundary artifacts as learnable forensic signals and uses five block types, no pooling in shallow and mid layers, residual learning, local feature fusion, and an ensemble over local patches. On the mixed three-class task—original, seam insertion, seam removal—it reports $\mathrm{Fun}_{\mathcal C}(M_i,M_{j_g})$9 accuracy with a single crop and $D \subset M$0 with ensemble size $D \subset M$1; its AUC values are $D \subset M$2, $D \subset M$3, and $D \subset M$4 for the three classes. Detection is consistently easier for insertion than for removal, which the paper attributes to stronger local correlations from duplicated pixels (Nam et al., 2020).

3. Fault-tolerant quantum computation and topological seam endpoints

In defect-free lattice surgery, a merge activates a uniform strip of local seam checks whose product equals the requested joint logical parity, such as $D \subset M$5 for a $D \subset M$6-merge or $D \subset M$7 for an $D \subset M$8-merge. On defect-adapted patches, this picture fails for three coupled reasons: boundary deformation and missing data support destroy or shift native seam checks; disabled checks and ancilla defects remove direct access to specific checks; and gauge-inferred super-stabilizers distribute the needed parity across schedule-tagged raw gauge outcomes. The problem is therefore formulated as a certified synthesis task: given two defect-adapted patches and a requested parity, determine whether that parity lies in the $D \subset M$9 row space generated by the admissible effective seam family $B$0 and the separated pre-merge constraints $B$1. The synthesis condition is

$B$2

and once $B$3 is found, the executable raw-bit selector is

$B$4

This separates parity-synthesis failure from patch invalidity. Circuit-level sampling under SI1000-MR with i.i.d. defects reports compile-yield improvements from $B$5–$B$6 to $B$7–$B$8 at defect rate $B$9, average effective-distance ratio $M$0–$M$1 versus $M$2–$M$3 for the baseline, and success-conditioned logical-error-rate overhead $M$4–$M$5 relative to the defect-free merge reference at $M$6 (Min et al., 28 Apr 2026).

A related seam-endpoint problem appears in topological quantum computation with gapped boundaries and boundary defects. In the doubled setting $M$7, a stable gapped boundary is modeled by a Lagrangian algebra, and defects between two boundary types are captured by categories such as

$M$8

When a seam carries a symmetry $M$9, crossed condensation maps bulk $B|_{M-D} \cong A$0-defects in $B|_{M-D} \cong A$1 to seam–boundary junction defects in $B|_{M-D} \cong A$2, where $B|_{M-D} \cong A$3 is determined by the braided auto-equivalence $B|_{M-D} \cong A$4. Fusion is given by relative tensor product over the intermediate boundary type, and projective braiding is inherited from the bulk $B|_{M-D} \cong A$5-crossed theory. In $B|_{M-D} \cong A$6, the $B|_{M-D} \cong A$7–$B|_{M-D} \cong A$8 seam terminating on a boundary supports a defect of quantum dimension $B|_{M-D} \cong A$9 with Ising-like fusion $1<2<3$0 (Cong et al., 2017).

4. Boundary seams in statistical mechanics, integrable systems, and boundary criticality

For logarithmic minimal models $1<2<3$1 on a strip, the seam-boundary defect problem asks how seams fused to a Robin boundary parameterize defect content in the bulk and determine the conformal spectra and boundary free energies. The Robin vacuum boundary is labeled by $1<2<3$2, and general $1<2<3$3 Robin boundary conditions are constructed by fusing an $1<2<3$4-type seam of width $1<2<3$5 and an $1<2<3$6-type seam of width $1<2<3$7 on top of that vacuum. The $1<2<3$8-type seam introduces precisely $1<2<3$9 defects into the bulk, and these defects become a good quantum number. With the special choice $$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$0, which enforces the drop-down property for boundary arcs, the resulting conformal weights are

$$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$1

Finite-size analysis was carried out up to $$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$2, with parity restriction $$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$3 for $$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$4 (Bourgine et al., 2016).

Morin-Duchesne, Ridout and Rasmussen introduced the boundary seam algebras $$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$5 to encode such seam-modified boundary sectors algebraically. Their admissible defect labels are

$$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$6

and the standard modules $$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$7 have dimension

$$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$8

At roots of unity, non-semisimplicity is organized by reflection across critical integers $$E(\ell)=\sum_{p\in\Omega} D_p(\ell_p)+\sum_{(p,q)\in\mathcal N} V_{p,q}(\ell_p,\ell_q),$$9 with $s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$0, leading to non-split short exact sequences

$s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$1

These sequences make the seam-boundary defect problem an explicit representation-theoretic problem for radicals, irreducibles, and projective covers (Langlois-Rémillard et al., 2019).

In classical integrable field theory on junction networks, the seam-boundary defect problem is formulated as the placement of purely transmitting defects at a junction of one-dimensional domains. For type I sewing at an $s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$2-branch junction,

$s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$3

and momentum conservation imposes

$s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$4

Because no antisymmetric $s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$5 can satisfy this for odd $s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$6, an even number of branches is mandatory. The paper therefore advises against terminating a purely transmitting seam on a single boundary, and instead treats a boundary termination as an even-branch junction or introduces auxiliary type II degrees of freedom (Corrigan et al., 2020).

A boundary defect in the quantum $s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$7 chain yields a different but related boundary-critical phenomenon. With

$s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$8

Jordan–Wigner and Majorana variables show that the impurity spin fractionalizes into two Majorana fermions, of which one decouples while the other couples to the bulk. The resulting boundary behavior realizes two-channel Kondo physics, with residual impurity entropy $s(\Delta)=\frac{1}{1+\exp(-\alpha(\Delta-\beta))},$9, a $s$00-function equal to $s$01, and a boundary-bound mode localized for $s$02 (Tang et al., 27 Jan 2025).

5. Continuum, atomistic, and free-boundary mechanics

In atomistic simulations of crystal defects, the outer computational boundary acts as a seam that interrupts the long-range elastic field generated by the defect. If the boundary data do not reproduce the true far-field equilibrium, the seam creates spurious tractions and image interactions. The variational analysis of finite-domain approximations derives sharp rates for several boundary-condition models. For point defects, Dirichlet truncation and periodic supercells satisfy

$s$03

whereas linear-elasticity far-field boundary conditions improve these to

$s$04

For dislocations, the improvement is more modest, with $s$05 and $s$06 (Ehrlacher et al., 2013).

For shallow Föppl–von Kármán surfaces, seam-localized line defects and metric anomalies enter through incompatibility relations for membrane and bending strains. An in-surface edge dislocation along a seam $s$07 is modeled by a line Dirac source such as $s$08, which produces the defect-driven plate system

$s$09

The resulting deformation folds about the seam with slope proportional to $s$10. More generally, wedge disclinations, twist disclinations, growth strains $s$11, and bending-growth fields $s$12 appear as explicit sources in the inhomogeneous FvK equations (Singh et al., 2021).

A three-dimensional field formulation based on Toupin’s theory of gradient elasticity represents point and line defects as singular distributions in a variational weak form. For a line defect on $s$13, the defect contribution is

$s$14

with dipole strength $s$15. The free energy

$s$16

introduces an internal length scale $s$17 that regularizes defect cores. The same framework models low-angle tilt boundaries as arrays of edge dislocations obeying the classical small-angle relation $s$18 (Wang et al., 2015).

In a Bernoulli free-boundary problem with compact inhomogeneities arrayed periodically, the seam is the hyperplane $s$19, and the defect problem becomes one of contact-line pinning. For rational $s$20, the advancing and receding pinned slopes satisfy

$s$21

$s$22

with $s$23 and $s$24 for $s$25. The width of the pinning interval is therefore governed by defect size, lattice density along the seam, and single-defect capacity-like coefficients (Feldman et al., 11 Dec 2025).

An auxiliary spectral interpretation of Green’s functions gives a wave-theoretic version of the seam-boundary defect problem. For Helmholtz operators, the Green’s function is the only physical eigenstate of an auxiliary problem with a $s$26-function potential at the source,

$s$27

and

$s$28

For a line seam $s$29, the field is continuous across $s$30 while the normal derivative jumps by $s$31. The paper emphasizes open and non-Hermitian settings and relates the local minimum of the Green’s function at the source to a chiral edge state circumventing a defect (Rivero et al., 2021).

6. Categorical and factorization-algebra formulations

A finite-group Dijkgraaf–Witten treatment of an embedded Seifert surface $s$32 with boundary knot $s$33 formulates the seam-boundary defect problem on the filtration $s$34. The initial data are finite three-object categories, or $s$35-parcels, given by a surjective conservative functor $s$36, where the objects $s$37 correspond to line, seam, and bulk. A coloring of a flag-like triangulation is a functor $s$38 with $s$39, and the untwisted state sum counts such colorings up to normalization by the internal gauge groups $s$40. The twisted theory assigns a partial $s$41-cocycle $s$42 on composable triples ending in the bulk object $s$43, and the invariant is

$s$44

The partially degenerate cocycle identities are designed exactly so that the state sum survives the flag-like extended Pachner moves involving the seam–boundary junction (Lee et al., 2015).

Factorization-algebra methods give a parallel formulation in perturbative field theory. A defect along $s$45 for a factorization algebra $s$46 on $s$47 is, by definition, a factorization algebra $s$48 on $s$49 together with an isomorphism $s$50. The geometric implementation is to blow up $s$51, replace it by a boundary, choose a local boundary condition

$s$52

that is Lagrangian for the $s$53-shifted local symplectic form on boundary jets, form the homotopy fiber product of solution sheaves, and push the resulting factorization algebra of observables back to $s$54. When the theory is topological normal to the defect, Rabinovich’s construction yields a genuine classical defect and, when a BV quantization exists, a genuine quantum defect. In codimension one, the seam behaves as a bimodule between the left and right bulk factorization algebras; in the presence of a physical boundary, the same formalism produces boundary modules and coherent seam–boundary junction data (Contreras et al., 2022).

These two abstract languages make explicit what many of the more concrete examples only imply: the seam-boundary defect problem is fundamentally a problem of extending local rules across a stratified singular set. Whether the ambient theory is combinatorial, categorical, variational, or algorithmic, the decisive step is the same—replace the broken local seam rule by a certified extension built from admissible effective data.