Improved accuracy traditional clothes pattern retrieval using VGG19 and Distance Metrics

Abstract

Research on traditional clothes pattern retrieval is still relatively new and still rarely done by researchers. The characteristics of traditional clothes patterns are so diverse that it requires constant methods and algorithms to carry out this research. In addition, the clothes pattern contains the characteristics of texture and shape. Traditional clothes can only have 1 pattern and some clothes have more than 2 patterns. For this reason, traditional clothes pattern retrieval methods and algorithms are needed that can increase the accuracy in clothes pattern retrieval.. In this study for feature extraction using VGG 19 with 4 distance metrics models, namely Manhattan, Euclidean, Chebyshev and Minkowski order 3. The average accuracy of retrieval of 74 traditional clothes patterns from 22 provinces is the highest in the Manhattan model with an average value of 87.04%, followed by the Euclidean model of 86.41%, Minkowski of 85.26% and the Chebyshev model with an average value of 66.07%. Even in Traditional clothes patterns can be recognized up to 100 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> by Manhattan and Euclidean methods. However, this study has not been able to overcome the Invariant dilemma of traditional clothes. For this reason, the continuation of this research still needs to be carried out.