CBCRS: an Open Case-Based Color Recommendation System
Yan Hong,Xianyi Zeng,Yuyang Wang,Pascal Bruniaux,Yan Chen
DOI: https://doi.org/10.1016/j.knosys.2017.11.014
IF: 8.139
2018-01-01
Knowledge-Based Systems
Abstract:In this paper, a case-based color recommendation system (CBCRS) is proposed for online color ranges (CRs) recommendation. This system can help designers and consumers to obtain the most appropriate CR of consumer-products (e.g., garments, cars, architecture, furniture …) based on the color image perceptual data of each specific user. The proposed system is an open system, permitting to dynamically integrate new CRs by progressively learning from users’ and designers’ perceptual data. For this purpose, a Color Image Space (CIS) is initially established by using Basic Color Sensory Attributes (BCSAs) to obtain the color image perceptual data of both designers and consumers. Emotional Color Image Words (CIWs) representing CRs are measured in the proposed CIS through a knowledge-based Kansei evaluation process performed by designers using fuzzy aggregation operators and fuzzy similarity measurement tools. Using this method, new CIWs and related CRs from open resources (such as new color trends) can be integrated into the system. In a new recommendation, user's color image perceptual data measured in the proposed CIS regarding different BCSAs will be compared with those of CIWs previously defined in the system in order to recommend new CRs. CBCRS is an adaptive system, i.e. satisfied CRs will be further retained in a Successful Cases Database (SCD) so as to adapt recommended CRs to new consumers, who have similar user profiles. The general working process of the proposed system is based on case-based learning. Through repeated interactions with the proposed system by performing the cycle of Recommendation – Display - Evaluation – SCD adjustment, users (consumer or designer) will obtain satisfied CRs. Meanwhile, the quality of the SCD can be improved by integrating new recommendation cases. The proposed recommendation system is capable of dynamically generating new CIWs, CRs and new cases based on open resources.