Premium
Classification of Persian carpet patterns based on quantitative aesthetic‐related features
Author(s) -
Soleymanian Moghadam Tayebe,
Ghanbar Afjeh Mansoureh,
Amirshahi Seyed Hossein
Publication year - 2021
Publication title -
color research and application
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.393
H-Index - 62
eISSN - 1520-6378
pISSN - 0361-2317
DOI - 10.1002/col.22555
Subject(s) - persian , histogram , pyramid (geometry) , orientation (vector space) , similarity (geometry) , pattern recognition (psychology) , artificial intelligence , categorization , measure (data warehouse) , set (abstract data type) , anisotropy , mathematics , computer science , image (mathematics) , data mining , geometry , linguistics , physics , philosophy , quantum mechanics , programming language
Abstract In this work, the extraction of significant features of Persian carpet patterns was studied. Four aesthetic related features were extracted for a collection of Persian carpet images. To this purpose, a set of 134 color images of three different categories of traditional Persian designs, named “Afshan,” “Lachak Toranj,” and “Torkaman” were collected. At first, the PHOG (Pyramid of Histogram of Orientation Gradients) measure was derived for all patterns to calculate complexity, anisotropy, self‐similarity, and Birkhoff‐like features. Based on the results, anisotropy and Birkhoff‐like features significantly categorize three carpet designs. According to the results, the combination of anisotropy and Birkhoff‐like features increases the accuracy of classification of samples to 97%.