Autoencoder Image Compression For High Compression And Acceptable Quality
Arnav Tyagi, Harshvardhan Aditya, Rishabh Khandelwal, Jagendra Pratap Singh, Yogesh Pal, Garima Jaiswal · 2023
This This article introduces a novel approach to image compression through the utilization of autoencoders, a class of neural networks adept at learning to distill an image's essential attributes and compactly represent them. The proposed technique entails training an autoencoder on a substantial image dataset and subsequently employing it to compress new images by encoding them into a lower-dimensional representation, which can be later decoded to reconstruct the original image. The paper conducts a comprehensive exploration of the method's compression effectiveness and image fidelity, offering comparative assessments with established compression methodologies. The findings underscore the approach's ability to achieve significant compression ratios while upholding commendable image quality, even surpassing established techniques in specific scenarios. Moreover, the study addresses the constraints and trade-offs associated with employing autoencoders for image compression, emphasizing the delicate balance between compression rate and image quality. It also outlines potential avenues for future research and development. In summary, this conference article not only underscores the utility of autoencoders for image compression but also accentuates their potential to enhance image storage and transmission efficiency across a range of applications.