PrismAlign: Prior-Steered Multi-View VLM Alignment for Hallucination-Robust Table OCR

Guangyi Liu; Qianjun Huang; Boyu Hou

cs.CVSep 18, 2026

PrismAlign: Prior-Steered Multi-View VLM Alignment for Hallucination-Robust Table OCR

Authors: Guangyi Liu, Qianjun Huang, Boyu Hou

Organizations: Huawei Technologies, Co., Ltd

Abstract

Table extraction suffers from frequent structural errors and semantic hallucinations. We propose PrismAlign, a multi-VLM framework aligning diverse visual perspectives to resolve ambiguity. It integrates priors of table logic to assess output plausibility, decoupling structural alignment from cell content alignment. A Bayesian decision strategy maximizes alignment accuracy by exploiting the correlation between extraction errors and computable rule violations. Evaluated on open-source and custom VLMs, PrismAlign reduces hallucinations and achieves state-of-the-art performance on OmniDocBench 1.5, as well as on the table category of CC-OCR and PureDocBench.

Figures & tables

VLM	Size	Overall↑	TextEdit↓	FormulaCDM↑	TableTEDS↑	TableTEDS-S↑
Nanonet 2 Mandal et al. (2025)	3B	83.13	0.114	78.02	82.77	87.75
MinerU 2.5	1.2B	90.67	0.047	88.46	88.22	92.38
dots.ocr Li et al. (2025a)	3B	88.41	0.048	83.22	86.78	90.62
MonkeyOCR-3B Li et al. (2025b)	3B	87.13	0.075	87.45	81.39	85.92
Deepseek-OCR	3B	87.01	0.073	83.37	84.97	88.80
Youtu-Parsing	2.4B	93.22	0.045	93.19	91.15	94.53

Table 1: Comparison on OmniDocBench 1.5

VLM	TableTEDS↑
MinerU 2.5	67.96
PaddleOCR-VL 1.5	81.51
Youtu-Parsing	81.37
PrismAlign	82.87

Table 2: CC-OCR table result

VLM	TableTEDS↑
MinerU 2.5	73.6
PaddleOCR-VL 1.5	75.6
FD-RL	80.50
PrismAlign	82.83

Table 3: PureDocBench table result

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

No.	Table structure rules
1	The output must contain valid HTML table control tags, e.g., <table>.
2	All <td> , <th> , and <tr> tags must be properly closed.
3	Column count must remain consistent across all rows, with column span attribute taken into account to ensure structural alignment.
4	Every column must contain the same number of rows, accounting for row span attribute where applicable.
5	HTML tags not related to tables should be filtered at the table structure level.
6	Rowspan and colspan attributes should not exceed the table’s dimensions, avoiding physically impossible layouts.

Appendix

Table 4: A set of table structure rules

No.	Name	Table semantic rules
1	Scale Constraints	The dimensions of the table must be bounded by a predefined threshold, e.g., 100 rows or 100 columns. Additionally, the values of rowspan and colspan attributes must not exceed this limit to prevent structural anomalies.
2	Content Repetition	Cells containing repeated n-grams, specifically, sequences of 64 tokens or more, are considered degenerate and should be discarded.
3	Boundary Validity	Leading or trailing rows or columns that are entirely empty are semantically void and should be removed.
4	Header Significance	A column consisting exclusively of numeric values must be anchored by a non-numeric header to provide necessary context.
5	Row Context	Rows primarily composed of numeric data must contain a categorical identifier (e.g., a label or key) in the first column.
6	Formula Integrity	Cells containing structurally malformed mathematical formulas, e.g., unclosed LaTeX delimiters, are not preferred.

Appendix

Table 5: A set of computable table semantic rules

Figure 2: PrismAlign component result comparison

Document Information Extraction

Document Information Extraction turns text, tables, forms, and images into structured, searchable data.

See all 290 papers

TableSeq: Unified Generation of Structure, Content, and Layout
Apr 17, 2026Laziz Hamdi, Amine Tamasna, Pascal Boisson +1Table Structure Recognition Multi-Token Prediction
LITIS, Rouen, Normandy, France · Malakoff Humanis, Paris, France
TABVERSE: Benchmarking Cross-Format Table Understanding in LLMs and VLMs
Jun 8, 2026Momina Ahsan, Sarfraz Ahmad, Ming Shan Hee +2Document Understanding Table QA
1Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) · 2Singapore University of Technology and Design (SUTD)
From Diagnosis to Correction: Benchmarking and Improving Real-World Table Parsing
Aug 10, 2026Jutao Xiao, Yuan Qu, Dongsheng Ma +7Table Structure Recognition Document Image Understanding
Zhejiang University · Shanghai Artificial Intelligence Laboratory · 1Zhejiang University, 2Shanghai Artificial Intelligence Laboratory +3

Apr 17, 2026J/K move · Enter open · S save

TableSeq: Unified Generation of Structure, Content, and Layout

Laziz Hamdi, Amine Tamasna, Pascal Boisson, Thierry Paquet

LITIS, Rouen, Normandy, France · Malakoff Humanis, Paris, France

Table Structure Recognition Multi-Token Prediction Document Image Understanding

Open paper AskPDF arXiv

We present TableSeq, an image-only, end-to-end framework for joint table structure recognition, content recognition, and cell localization. The model formulates these tasks as a single sequence-generation problem: one decoder produces an interleaved stream of HTML tags, cell text, and discretized coordinate tokens, thereby aligning logical structure, textual content, and cell geometry within a unified autoregressive sequence. This design avoids external OCR, auxiliary decoders, and complex multi-stage post-processing. TableSeq combines a lightweight high-resolution FCN-H16 encoder with a minimal structure-prior head and a single-layer transformer encoder, yielding a compact architecture that remains effective on challenging layouts. Across standard benchmarks, TableSeq achieves competitive or state-of-the-art results while preserving architectural simplicity. It reaches 95.23 TEDS / 96.83 S-TEDS on PubTabNet, 97.45 TEDS / 98.69 S-TEDS on FinTabNet, and 99.79 / 99.54 / 99.66 precision / recall / F1 on SciTSR under the CAR protocol, while remaining competitive on PubTables-1M under GriTS. Beyond TSR/TCR, the same sequence interface generalizes to index-based table querying without task-specific heads, achieving the best IRDR score and competitive ICDR/ICR performance. We also study multi-token prediction for faster blockwise decoding and show that it reduces inference latency with only limited accuracy degradation. Overall, TableSeq provides a practical and reproducible single-stream baseline for unified table recognition, and the source code will be made publicly available at https://github.com/hamdilaziz/TableSeq.

Abstract

Document Information Extraction

Explore similar work

TableSeq: Unified Generation of Structure, Content, and Layout

TABVERSE: Benchmarking Cross-Format Table Understanding in LLMs and VLMs

From Diagnosis to Correction: Benchmarking and Improving Real-World Table Parsing