JPEG is one of the most popular image compression techniques in the world. Its effectiveness has led to it being used in diverse sectors such as satellite imaging, medical imaging, image storage systems and multimedia. With the diverse use of JPEG compression algorithms, it has also become necessary to develop deblocking algorithms to mitigate the compression loss caused by the compression. With the advent of deep learning, several
methods have been developed for JPEG image deblocking. The quality factor or QF value is vital to the compression process. Most deep JPEG deblocking networks face the challenge of requiring the image to be compressed by a QF value which is part of the training
process. If the image is compressed by any other QF value, the performance and deblocking quality of the network severely degrades.
In this thesis, two different schemes are proposed to solve this issue. The first proposed scheme aims to tackle the problem from the out-of-distribution point of view, whereas the
second proposed network aims to tackle the problem from a meta-learning point of view. The effectiveness of the proposed schemes is validated by conducting experiments employing two different benchmark datasets. The proposed networks are shown to outperform the state-of-the-art deep JPEG deblocking networks as shown by the quantitative and qualitative comparative studies.