PROJECT TITLE :

Deep Learning-Based Picture-Wise Just Noticeable Distortion Prediction Model for Image Compression

ABSTRACT:

Having a good eye for a picture An Image Processing technique known as Just Noticeable Difference, or PW-JND, is commonly employed in perception-oriented image and video processing. Conventional JND models, on the other hand, calculate each pixel or sub-JND band's threshold individually, which may not adequately reflect the whole masking impact of a picture. Deep learning-based PW-JND prediction model for image compression is proposed in this paper. Prior to discussing the framework, we define PW-JND as a multiclass classification issue, and offer a method to transform the multi-class classification problem into a binary classifier-solved classifier problem. Perceptually lossy/lossless predictor is a deep learning-based binary classifier that can tell if a picture is perceptually lossy or not. Sliding window search approach is proposed for predicting PW-JND using the perceptually lossy/lossless predictor's prediction findings. An average of 0.79 dB of absolute prediction error was found in the suggested PW-JND model, which shows that it is superior to traditional JND models in terms of perceptually lossy/lossless prediction accuracy.