Approximation of CIEDE2000 color closeness function using Neuro-Fuzzy networks

The development of efficient algorithms that perform qualitative operations in the processing and segmentation of graphical images requires good knowledge on both technical and biological aspects of the vision. Due to the subjective nature of the human vision, the evaluation of color identity and closeness, that has been widely used in a variety of image processing problems, is still subjected to research and improvement. This paper analyzes the problem of color tolerance, overviewing past and existing solutions, and suggests a simple Adaptive Neuro-Fuzzy Inference System (ANFIS)-based model as an alternative to the formula-based closeness evaluation. The aim of this research was to develop an alternative to the unified closeness formula, which could be built on expert knowledge or be adapted to fit a particular target. Methods of sample selection, training approaches, and configuration of the ANFIS network were used and a comparative analysis between the ANFIS models and the neural networks was provided. Additionally, the efficiency of regular back-propagation, hybrid, and stochastic learning methods were evaluated along with a critical analysis on the widely used hybrid training method. Experiments were designed and ran on a Java-based platform-independent system, developed as part of our research, to provide flexibility in the modeling and integration for the ANFIS based tasks.

Automated summary

The paper analyzes the problem of color tolerance and proposes an ANFIS-based model as an alternative. The aim of the research was to develop an alternative to the unified closeness formula that could be built on expert knowledge or adapted to fit a particular target.