In this project, image compression utilizing visual redundancy is being investigated. Inspired by recent advancements in image compression techniques, various contemporary image compression frameworks are compared towards visual quality rather than pixel-wise fidelity. An original image is analyzed at the encoder side so that portions of the image are intentionally and automatically skipped for different techniques. The parameters are determined to compare the effectiveness of the techniques. Performance of different transform techniques are analyzed and later, image formats. The merits and demerits of each technique are pointed out and suggestions for improvement are provided.
Here the effectiveness of the following transform techniques are compared:
- Discrete Cosine Transform (DCT) Algorithm
- Haar Wavelet Transform Algorithm
- Karhunen-Loeve Transform (KLT) Algorithm
This project also includes comparing Better Portable Graphics (BPG) Image format and JPEG. Given the importance of knowing the updated image compression techniques, it is required to have a sound understanding of the behavior of different compression algorithms.
It is found out that BPG proves to be a better format compared to JPEG. Its purpose is to be a more compression-efficient replacement for the JPEG image format when quality or file size is an issue. DCT proves to be the optimum transform for the image. KL is the best comparing the MSE and PSNR of each level of compression, followed by DCT and Haar. But considering the computation time and certain drawbacks of KL, it is established that DCT is the best transform for image compression. Like the Fast Fourier Transforms, the algorithm has to be found out to compute KL Transform faster, such that the transmission rate could be increased and our aim for compression towards visual quality rather than pixel-wise fidelity could be achieved.