Scholarlink

<p>The acquisition and processing of digital images and videos has become increasingly common in modern times owing to the expansion of internet services and their associated portable devices. With the limited processing capacity and memory constraints of such devices, the need for innovative compression algorithms emerges. In contemporary research, Researchers have widely adopted Deep Learning (DL) models to enhance image and video compression techniques. Despite the wide usage, the performance and memory constraints of analytical structures are predominant with the DL models that are affected by loss estimation, pruning, and memory localization. So, to analyse the estimation and minimization of memory features in DL models, an Iterative Predictive Transform (IPA) and an Intuitive Prediction Algorithm (IPT) technique have been proposed, facilitating the reduction of compression loss with improved loss punning for DL methods. The proposed transformation and prediction algorithm optimizes weight parameters by using a log sigmoid function to lower the error rate and provide superior accuracy for visual compression. The efficiency of the proposed compression techniques has been evaluated using several performance metrics, including Structural Similarity Index Measure (SSIM), Mean Square Error (MSE), Peak Signal-to-Noise Ratio (PSNR), Compression Ratio (CR), and classification accuracy. Experimental results indicate that the IPA-IPT-based methods specifically IPA-AE, IPA-DL, and IPT-Convolutional Neural Network (CNN) models—achieved an exceptional accuracy of 99.998%, with compression and multiplication factors of 17 and 10, respectively. These outcomes demonstrate that the proposed approaches significantly outperform existing state-of-the-art algorithms based on Generative Adversarial Networks (GANs).</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An intuitive IPT-IPA based deep learning approach for video compression techniques

  • 期刊论文

  • 发表日期：

    2025年10月1日

查看全文
  • Mudhavath Ramesh Naik,
  • Jayendra Kumar,
  • Arvind R. Yadav

摘要

The acquisition and processing of digital images and videos has become increasingly common in modern times owing to the expansion of internet services and their associated portable devices. With the limited processing capacity and memory constraints of such devices, the need for innovative compression algorithms emerges. In contemporary research, Researchers have widely adopted Deep Learning (DL) models to enhance image and video compression techniques. Despite the wide usage, the performance and memory constraints of analytical structures are predominant with the DL models that are affected by loss estimation, pruning, and memory localization. So, to analyse the estimation and minimization of memory features in DL models, an Iterative Predictive Transform (IPA) and an Intuitive Prediction Algorithm (IPT) technique have been proposed, facilitating the reduction of compression loss with improved loss punning for DL methods. The proposed transformation and prediction algorithm optimizes weight parameters by using a log sigmoid function to lower the error rate and provide superior accuracy for visual compression. The efficiency of the proposed compression techniques has been evaluated using several performance metrics, including Structural Similarity Index Measure (SSIM), Mean Square Error (MSE), Peak Signal-to-Noise Ratio (PSNR), Compression Ratio (CR), and classification accuracy. Experimental results indicate that the IPA-IPT-based methods specifically IPA-AE, IPA-DL, and IPT-Convolutional Neural Network (CNN) models—achieved an exceptional accuracy of 99.998%, with compression and multiplication factors of 17 and 10, respectively. These outcomes demonstrate that the proposed approaches significantly outperform existing state-of-the-art algorithms based on Generative Adversarial Networks (GANs).