Research On Image Lossless Compression Based On Distributed Coding

With the rapid development of communication technology, multimedia technology, cloud compute and big data et al., multimedia applications have been pervasive in people’s daily lives, and higher image quality is required in more aspects of life. At the same time, with the development and popularization of wireless networks and mobile devices, the resource limited encoder and the unstable channel quality put forward new challenge on image coding. The traditional image coding methods, such as JPEG, JPEG2000, JPEG-LS and the intra prediction methods in H.26x, explore the spatial correlation at the encoder, and transmit the source coded stream with channel coding to protect the transmitted information, which make the encoder with high computational complexity and poor robustness. Therefore, the traditional image coding methods cannot satisfy the new application requirements.The researches on distributed coding theory brought new strategy for these new applications under wireless environment. According to Slepian-Wolf theory and Wyner-Ziv theory, the correlated sources can be compressed by separate encoders without perforamence loss as long as they are jointly decoded. Distributed coding theroies laid the solid foundation for development of distributed image coding techniques. Distributed coding removes the complex decorrelation processes into the decoder, which makes the scheme can make full use of the resources at the decoder. Therefore, distributed coding has the feature of low encoding complexity, small resource consumption and high robustness. This technique has an important application value for onboard image coding, and it has become a hot research area. Besides, with the development of cloud compute and big data in recent years, it brings the new opportunity for distributed coding. Because of the limited resources at the encoder while the sufficient resources at the decoder (there are a lot of correlated data at the cloud), the asymmetric phenomena of resource allocation makes the distributed coding be suitable for this application.This dissertation designs the lossless distributed image coding scheme for hyperspectral image coding, multiview remote sensing image coding and the cloud based image coding, aiming to design an image coding scheme with simple encoder, low storage requirement, high coding efficiency and high robustness. The main innovations and contributions of this dissertation are listed as follows.(1) This dissertation proposes a distributed coding scheme for hyperspectral image lossless coding with region-based adaptive prediction strategy. Firstly, since hyperspectral images have hundreds of bands with high resolution, however, all the bands contain the context of the same place, this scheme designs an image segmentation algorithm to classify the image into several classes, and then designs a region based prediction method for each class. Secondly, since different bands of the hyperspectral images are interrupted by different noise, this scheme designs an algorithm to adaptively choose the reference bands. At last, this scheme proposes to combine the MRF model with LDPC decoder, such that the spatial correlation can be removed, and the performance would be improved. In summary, this scheme gets the high compression performance with simple encoder by removing both the spatial and spectral correlation at the decoder side.(2) This dissertation proposes a distributed multiview remote sensing image coding scheme with line-based prediction. This scheme solves the problem of the limited computing and storage resources at the same time. In the proposed scheme, it first classifies the pixels into different types, and then designs the line-based adaptive filters for different types to remove the intra-view correlation, and designs the adaptive template matching techniques to remove the inter-view correlation. Therefore, this scheme has the advantages of simple encoder, low storage requirement and high compression efficiency.(3) This dissertation proposes a cloud-based distributed image coding scheme. This scheme sends a downsampled image into the decoder, and the decoder gets the side information by the downsampled image and the plentiful correlated images in the cloud. Besides, this scheme designs the down sample and up sample filters to improve the image quality, and then helps the decoder get better side information. This scheme changed the condition that the traditional coding methods can only explore the spatial correlation at encoder, and explores the correlation between images by taking advantage of the correlated images in the cloud. Therefore, this scheme achieves a better performance compared with the traditional methods.