An Aquatic Image Compression Scheme based on Optimized Deep Convolutional Autoencoder

Nowadays, in daily life, several online platforms have generated an enormous amount of data, mainly images, audio, and video. Uncompressed multimedia information, such as images, graphics, audio, animation, and video, needs massive storage space and transmission bandwidth. Suitable image compression methods to overcome excessive data traffic are necessary. The image compression scheme reduces image size, which is beneficial for storage and transmission. Multiple techniques have been utilized to solve the difficulties, but most suffer a significant drawback: important information must be recovered in the reconstructed image. An optimized deep convolutional autoencoder model has been implemented using the deep learning approach for solving the challenges. The proposed model has many layers and filters to develop an effective, efficient image compression method. The unsupervised machine learning approach compresses the image using the backpropagation technique and finally reconstructs the image with minimum information loss. One new instance has been incorporated into the architecture to improve image compression performance. The methodology performed better at the time of image compression. Due to this problem, we select convolutional neural networks, followed by generative adversarial networks, as a solution to reduce diverse compression artifacts. This research covers the compression of underwater images based on a deep convolutional autoencoder. The concept of underwater image acquisition techniques and their analysis are also discussed. The proposed image compression approach is studied using performance parameters, like Space Saving (SS(%)), and PSNR is differentiated with state-of-the-art methods. Experimental outcomes indicate the proposed technique acquires higher SS (%) and PSNR, reduced space complexity, and better image quality than the existing image compression system. Using the Marine Animals dataset, the proposed model achieved SS (%) 83.33, PSNR 72.60 (dB), and Structural Similarity Index Measurement (SSIM) 0.9517 values. Also, the proposed model achieved SS (%) 83.33, PSNR 74.77 (dB), and SSIM 0.9766 values using the Sea Animal dataset. However, it has produced a new root in the future investigation for the improvement of the method, such as better performance factor of compression and minimization of data loss for the 3D image. © (2024), (International Association of Engineers). All Rights Reserved.