Study on Constructing a Categorical Color Perception Model for Selecting Colors Used in CAPTCHAs

Abstract

Completely automated public Turing test to tell computers and humans apart (CAPTCHA) distinguishes between human and machine access to web pages. However, CAPTCHAs that use visual characteristics have been studied to make machine responses more difficult. In this study, a model that reproduces human color perception was constructed using a neural network with the aim of utilizing the ambiguity of human color perception in CAPTCHA. The results of the investigation of training data showed that the data used to train the constructed color perception model was optimal when only colors that could be classified as categorical colors were used.