A Study On Image Coding And Transmission And Its Applications In Industry Monitoring

The key technology in the process of image coding and wireless transmission application is studied from two directions in this dissertation. On the one hand, the machine learning method is applied into image compression and the corresponding performance is discussed, through breaking down the frame of current image coding standards. On the other hand, the problem of blocking artifacts in existing block-based methods and the technologies of error resilient image coding and transmission are researched, in order to improve the performance of image coding and transmission over wireless channels. Part of the research results have been applied in the remote environment monitoring system. The main contents of this dissertation are outlined as follows,(1) Image data are encoded combining with weighted support vector machine (SVM) and discrete cosine transform (DCT) by applying the machine learning method into image compression. The weight function models for penalty factor are generated according to the energy distribution characteristic of DCT coefficients and the human visual system models, and then these models are applied into the learning process of weighted SVM. Different penalty factors are assigned to the sample data of different importance, and the tradeoff can be determined between training errors and model complexity. This algorithm adaptively chooses support vectors ranging from low frequency regions to high ones during the training process, rather than directly low-pass filtering sample data. In this way, some image details and edges can be preserved to improve the quality of the reconstructed image. Fixed insensitivity error has always been used by the proposed SVM-based strategies for image compression, but it can be adjusted to control approximation accuracy of data in the frequency domain. Based on the idea, a new image compression approach is proposed on the basis of support vector domain description and adaptively weighted SVM. On each segmented image subblock, its corresponding weight function model is obtained according to each DCT coefficient's distance to the center of the smallest enclosing hypersphere in the high dimension feature space. These established models are eventually embedded into the SVM scheme for image compression, in order to improve the compression performance.(2) The problem of blocking artifacts for block-based DCT compression meth- ods at low bit rates is studied. An adaptive DCT-domain algorithm for the reduction of blocking artifacts is proposed to improve the visual effect of reconstructed images. The algorithm is executed in the transform domain. The shifted block is obtained directly form its adjacent blocks via the decomposition and combination of DCT coefficients. Different post-processing methods are then adopted for different blocks based on block activity classification. The proposed algorithm can achieve satisfactory effect on reducing blocking artifacts and improve the objective and subjective image quality as compared to other post-processing techniques.(3) The problem of high sensitivity of compressed image data for transmission errors over wireless channels is investigated. Due to the fact that almost any error can not lead to acceptant effect on reconstructed images on the receiver, a simple and applied error resilient scheme is proposed based on a novel anti-jamming algorithm and a scattered packet algorithm for image data based on a window templet.Firstly, the anti-jamming algorithm is used to protect header information of image data according to the idea of unequal error protection. Secondly, the proposed packet algorithm, which scatters and packets compressed data blocks in certain order, decreases the loss probability of consecutive data blocks at the same time, and then makes image data more robust during the process of wireless transmission.(4) In the application of remote environment monitoring based on wireless mobile networks, not only transmission data needs to be decreased in order to accommodate their relatively narrow bandwidth and lower operation cost, but also image transmission are ensured to be stable and reliable. Aiming at these requirements, a real-time scene change detection algorithm is proposed. Change detection between image sequences is rapidly executed in the compressed domain, and then reliable scene change indication is provided for the wireless digital image monitoring system. This algorithm is well suited for real-time processing of common embedded devices due to its simplicity and availability. Based on the scene change detection algorithm and proposed anti-jamming algorithm, a highly error-resilient scheme for image coding and wireless transmission is designed and applied into the remote environment monitoring system, which achieves good practicability.