Publication: Edge sharpness detection via Image brightness adjustment with data mining insights
Loading...
Date
2024-07-12
Authors
Lim, Shiy Voon
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Visual inspection for maintaining product quality is critical, particularly in sectors like electronics and manufacturing where precision is essential. Machine vision systems have become crucial for reducing human error and labour costs in quality control processes. Existing research has extensively studied edge detection techniques such as Canny, Sobel, and Prewitt operators, each with unique strengths in noise suppression, accuracy, and computational efficiency. However, these methods face challenges in handling noise and varying light conditions, necessitating meticulous parameter tuning which can compromise performance. Additionally, there is limited research on the integration of classification models with edge detection outputs to enhance image analysis. Addressing these gaps, this study aims to develop a predictive classification model for edge sharpness and contrast, predict the optimal contrast for various brightness levels, and automate the determination of optimal light brightness using edge sharpness and contrast analysis. The study employed a multidisciplinary approach, integrating image processing techniques, mathematical and statistical concepts, and data mining strategies. A diverse dataset of images was captured under diverse lighting conditions, and a customized edge detection algorithm using Python was developed. Linear Regression, Multilayer Perceptron, and Gaussian Processes were trained and evaluated using 10-fold cross-validation in WEKA to predict the contrast values. Linear Regression algorithm returned the best predictive model for edge sharpness and contrast, achieving over 90% accuracy in predicting optimal contrast values. Additionally, a Python program was developed to automate light brightness determination, significantly enhancing edge detection accuracy under varying lighting conditions. This study’s impact lies in advancing edge detection techniques by integrating predictive classification models developed, contributing to more robust and reliable image processing methods for industrial applications.