Feature Extraction Techniques for Patterned Images: A Systematic Literature Review

This research aims to conduct a systematic review of feature extraction techniques in patterned images. A total of 47 journals were analyzed to identify the techniques used, their advantages and disadvantages, and the performance outcomes in various applications. The results indicate that statisticalbased techniques such as the Gray Level Co-occurrence Matrix (GLCM), transformation-based techniques like discrete wavelet transform (DWT), visual descriptors such as SURF, and deep learning techniques like CNN are widely used and show excellent performance. For instance, statistical techniques are used for their simplicity and effectiveness. Additionally, deep learning techniques have succeeded in automatically analyzing image types from various fields based on neural network architectures, particularly CNN. Recent trends also show that combining traditional methods with deep learning, such as GLCM + Autoencoder and Haralick + Autoencoder, achieves the highest results, with a performance of 99.89%. This research states that although each technique has its pros and cons, deep learning techniques tend to be the primary choice for complex and high-precision feature extraction. The integration of classical methods with deep learning technology offers significant potential for further advancements in feature extraction for patterned images.