Customers preferences are difficult to predict automatically because they are subjective and relative to human likeness patterns which are different from each people. This paper proposes the combination of Interactive Genetic Algorithm (IGA) with Artificial Neural Networks to model the fitness function and find the optimally colored image according to user's preference. As the search space for the IGA is very large, proposed system reduces it creating genotypes with colors restrictions developed with a new graph coloring technique. Two functions are proposed for coding color restrictions. The system presented in this study takes the case of the color design of a building for illustration purposes. Experimental results show that 80% reduction in the number of users evaluations is achieved with proposed combination and speedup in IGA evolution is achieved when color restrictions are coded by proposed mutation and crossover operators.
Mathematics Subject Classification (2000)68T20 · 68T05 · 68U10 · 68U20 · 90C35 · 90C59 · 91B42 · 92B20
Colors play an important role for customers in making decisions on what they like or dislike. Frequently, customers are overloaded by color combinations to consider and they may not have the time or knowledge to personally evaluate all these combinations in a product design. This paper proposes a color recommendation system which includes design concepts as rules constraining the interactive search made by genetic algorithms to follow customer preferences. As the search space is very large and it changes with contextual information, proposed system combines graph coloring techniques with artificial neural networks to keep color restrictions during system evolution and model the fitness function provided by customers. In order to illustrate an application of proposed system, building images are used as example. After including conceptual coloring rules for building images in the system by two different methods, a questionnaire study was used to verify which approach suggested better images according to customer preferences. Experiments demonstrate that only if contextual information is included in the learning process system predictions keep close to customer's evaluation.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.