The arithmetic code has come to be used for image coding about 40 years ago. It was first adopted in bi-level image coding standard and nowadays widely adopted in multi-level image coding and video coding standards. Arithmetic code is classified as a non-block code in the information theory, and quite powerful by its ability and robustness for various target images. In this paper, its largest advantage for coding multi-context sources is described by comparing with the case using block codes. Additional advantage of arithmetic coding is its robustness based on the affinity with statistic learning. Therefore, the separation of model and entropy coding can be easily done and the current context and the observed symbol are enough for applying an arithmetic coding, while code set design and selection rule from the plural codes will be required in Huffman coding. Also, there were various efforts to make the arithmetic codes practical by referring to the designing parameters of known code including MELCODE.

Precise automatic segmentation of lung organs by computed tomography (CT) is a prerequisite for organ identification, pathological localization and treatment of lung diseases. However, accurate lung segmentation remains a major challenge due to the shape, size and location of the lungs. In recent years, U-net network structures and variations have been applied to various medical image segmentation, but these networks still have limitations and shortcomings in solving the vanishing gradient problem and contextual semantic feature extraction. In this paper, we propose a deep residual network called ECA-Resunet to segment lung organs with an efficient attention mechanism. The structure of ECA-Resunet is similar to Res-UNet. It uses deep residual units to form the entire encoder-decoder network and adds an efficient channel attention mechanism to the encoder. Compared to other networks, the advantage of ECA-Resunet is that it uses a deep residual network, which makes it difficult for the gradient to vanish. The encoder introduces an efficient channel attention mechanism to increase the weight of key regions in order to highlight key features and better learn the semantic features of the image context. This way, we can design a network with fewer settings and achieve better semantic segmentation results without changing the original image size. The results show that evaluation metrics such as Miou Score, Dice Effective Score, and Sensitivity of ECA-Resunet outperform those of the comparison network.