Myhistory: analyzing your family photos by using deep learning and large language models

Abstract

Image analysis has become increasingly vital across domains such as healthcare, security, and entertainment, where visual data quality and interpretability play a critical role. This research investigates three core image analysis tasks: detecting image blurriness, identifying closed eyes in facial images, and classifying images into predefined categories. By combining traditional methods like Laplacian variance for blurriness detection and Haar cascades for open eye detection with state-of-the-art deep learning models such as MobileNetV2 and ResNet-50, a robust and integrated image processing pipeline was developed. The study utilized diverse datasets, incorporating images with varied sharpness levels, facial conditions, and object categories, and applied preprocessing techniques like normalization and data augmentation to enhance model performance. Experimental results revealed that deep learning models significantly outperformed traditional techniques in accuracy and robustness, particularly for image classification and Open eye detection under complex conditions. Despite these advancements, challenges such as generalization to real-world scenarios, computational demands, and environmental sensitivity persist. The research emphasizes the potential of hybrid approaches to overcome these limitations and provides actionable insights for future developments in image analysis systems. This study contributes to improving the efficiency and reliability of visual data processing, offering practical applications in industries reliant on image quality and classification.