Learning-driven lossy image compression: A comprehensive survey

In the field of image processing and computer vision (CV), machine learning (ML) architectures are widely used. Image compression problems can be solved using convolutional neural networks (CNNs). As a result bandwidth memory constraints, images is necessity. There three types information found in images: useful, redundant, irrelevant. this survey, we will discuss how ML used to compress lossy images. Firstly, describe background compression. Next, classify ML-based frameworks into subgroups based on their architectures. Auto-encoders (AEs), variational auto-encoders (VAEs), CNNs, recurrent (RNNs), long short-term memories (LSTMs), gated units (GRUs), generative adversarial (GANs), transformers, principal component analysis (PCA) fuzzy means clustering among these subgroups. By analyzing learning-driven frameworks, present pros cons each subgroup. Lastly, outline several research gaps future directions