A COMPUTATIONAL FRAMEWORK FOR EXPLAINABLE ASSISTED SIGNATURE SCREENING IN FORENSIC DOCUMENT ANALYSIS

Resumo

This paper refactors a previously forensic-oriented methodological manuscript into an applied-computing article suitable for the Iberoamerican Journal of Applied Computing. The proposed contribution is an explainable computational framework for assisted screening of questioned signatures from static images, with emphasis on feature transparency, multisample comparison, and decision support rather than automated attribution. The framework organizes the workflow into image quality assessment, pre-processing, extraction of geometric and texture descriptors, multisample normalization, divergence-aware scoring, and structured reporting. A case-based demonstration is presented using one questioned signature and four reference signatures, for which the original technical execution indicated mixed evidence: broad-form compatibility coexisted with divergence in slant, ink density, baseline behavior, projection entropy, and anti-remake indicators. Instead of overclaiming authorship, the framework treats intermediate outputs as explainable evidence layers and supports an inconclusive result when the observed pattern does not justify strong attribution. The article contributes a computational perspective that aligns signature examination with current concerns in applied machine learning and forensic decision support, namely explainability, calibrated uncertainty, and reproducible reporting. The manuscript also adds a real bibliography grounded in signature verification, forensic handwriting comparison, and explainable decision support, addressing a limitation identified in the previous submission history.