Publication:
Automatic emulation of human experts for estimating chromatic quality in digitization of graphical documents

Thumbnail Image
Identifiers
URI: https://hdl.handle.net/10016/23693
Files
automatic_2016.pdf (2.61 MB)
Publication date
2016
Defense date
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Google Scholar
Export
Research Projects
Organizational Units
Journal Issue
Abstract
This work aims to provide a critical examination of different approaches to creating models of automated quality control systems for digital images in digitization projects for photographic heritage collections. It investigates the feasibility of using machine-learning algorithms that work on sets of images previously evaluated by experts to obtain models on which to construct a high performance visual algorithm. We analyzed the data collected after conducting a psychometric experiment in which four human experts evaluated a set of three series of 300 degraded images by assigning each image to different quality classes. This analysis concludes that it is not possible to talk about commonly used simplistic models based on continuous acceptance ranges for colour metrics on an isolated basis, and therefore that it is necessary to investigate more complex models. This study demonstrates that a model based on a machine learning rule-based system employing the CIE 1976 or CIEDE 2000 metrics along with the hue, saturation and lightness colour perceptual attributes emulates the human image quality experts with a high degree of efficacy, above 85%, opening an interesting way to get higher performance visual algorithms to automatically evaluate image quality in the context of digitization of photographic collections.
Description
Traducción al inglés de una versión ampliada del artículo: Robledano-Arillo, J.; Moreno-Pelayo, V.; Pereira-Uzal, J.M. (2016). Aproximación experimental al uso de métricas objetivas para la estimación de calidad cromática en la digitalización de patrimonio documental gráfico. Revista Española de Documentación Científica, 39 (2): e128. doi: http://dx.doi.org/10.3989/redc.2016.2.1249
Keywords
Document digitization, Photography, Image quality assessment, Machine learning, C4.5 algorithm, Visual algorithms, Digitalización de documentos, Fotografía, Evaluación de calidad, Aprendizaje automático, Algoritmos visuales
Bibliographic citation
Collections
DBD - Artículos de Revistas
DI - KRG - Artículos de Revistas