Noticeable Distortion

Just noticeable distortion (JND) research focuses on identifying the minimal level of image or video distortion imperceptible to the human eye, crucial for optimizing compression algorithms. Current research emphasizes leveraging deep learning models, often incorporating semantic information or image quality assessment metrics, to predict and utilize JND for more efficient compression. These advancements lead to improved rate-distortion performance in image and video coding, resulting in smaller file sizes without significant perceived quality loss, with applications ranging from image compression to video streaming. The development of lightweight, adaptable JND models is a key area of ongoing investigation.