Property EnableNoiseRejection

Namespace

LMKit.Extraction.Ocr

Assembly

LM-Kit.NET.dll

EnableNoiseRejection

Gets or sets a value indicating whether recognized pages are screened after OCR and scored for the likelihood that their content is genuine text rather than recognition noise.

public bool EnableNoiseRejection { get; set; }

Property Value

bool

true to screen each page and publish the verdict through ContentConfidence; false to skip the screening entirely. The default is false.

Remarks

When OCR runs over content that carries no machine-readable text, such as a natural photograph or a texture, the recognition engine can produce word-shaped artifacts. Indexing or extracting that output pollutes downstream search and analysis results.

With this option enabled, each page is scored after recognition by fusing lexical plausibility, spatial coherence, and confidence distribution signals. The page is returned intact, with all recognized elements preserved, and carries the calibrated verdict in ContentConfidence; see that property for the recommended reject and review thresholds. The screening is pure arithmetic over the recognized words, adding only microseconds per page.

The scoring is deliberately conservative: any strong sign of genuine text yields a confident score, so ordinary documents, including low-quality scans and photographed receipts, are not affected.