International Journal of Science, Engineering and TechnologyAuthors: Aditya Ram N, Yashwanth Kumar G K, Senthil Kumar B
Abstract: In recent years, deep learning has made a substantial impact on computer vision systems, especially in image processing, feature extraction, and reconstruction. The traditional method using OpenCV with Keras-based convolutional neural networks (CNNs) has been widely employed for image analysis and classification tasks. However, such methods are often dependent on extensive manual preprocessing, require large amounts of labeled data, and involve substantial computational complexity. This paper proposes a comparative analysis of the performance of an OpenCV-Keras-based pipeline and an Autoencoder-based deep learning model for image reconstruction and representation learning. The OpenCV-Keras-based pipeline is based on a traditional supervised learning strategy, whereas the Autoencoder-based model uses an unsupervised learning strategy to learn compact representations of the input images automatically. The experimental analysis reveals that the Autoencoder-based model outperforms the OpenCV-Keras-based pipeline in terms of noise removal, feature retention, reconstruction accuracy, and computational complexity. The paper concludes that Autoencoders can serve as a more scalable and intelligent alternative to traditional OpenCV-Keras-based pipelines, especially in real-time applications. Prior studies in representation learning and reconstruction using autoencoders have demonstrated their effectiveness in noisy environments. [1], [2]
DOI:
