# Colour segmentation of printed fabrics by integrating adaptive neural network and density peak clustering algorithm

## TL;DR

A new method for color segmentation in printed fabrics combines adaptive neural networks and density peak clustering to improve accuracy and efficiency.

## Contribution

A novel color segmentation algorithm integrating SOM and DPC for printed fabrics, achieving higher accuracy and faster execution.

## Key findings

- The proposed method reduces execution time by nearly 40% compared to self-organized mapping networks.
- The average color difference (ΔE) after segmentation is as low as 0.7, outperforming other algorithms.
- The average color value reaches 87.49 with a standard deviation as low as 2.18, indicating high segmentation stability.

## Abstract

With the development of computer vision and image processing technology, color segmentation of printed fabrics has gradually become a key task in the textile industry. However, the existing methods often face the problems of low segmentation accuracy and poor computational efficiency when dealing with high complexity patterns and similar colors. To address the above problems, a new color segmentation algorithm for printed fabrics is proposed by integrating the self-organizing mapping network (SOM) in adaptive neural network and the density peak clustering algorithm. The method achieves topological mapping learning of color features through SOM, and then uses DPC for density-driven fine clustering division, which effectively improves the accuracy and stability of color segmentation. The experimental results show that the proposed method shortens the execution time by nearly 40% compared with the self-organized mapping network, and the average color difference (ΔE) of each region after color segmentation is as low as 0.7, which is significantly better than other algorithms. Meanwhile, in the detection of the four types of printed fabric samples, the obtained average color value is up to 87.49 (the higher the 0–100 score value indicates that the color is more significant), and the smallest standard deviation is 2.18 (the smaller the value indicates that the color segmentation is more centralized), which further verifies the comprehensive advantages of the algorithm in terms of segmentation accuracy and stability. In conclusion, the proposed method provides an effective reference for improving the quality and efficiency of color segmentation of printed fabrics.