Low Bit-rate And High-fidelity Compression Method For Remote Sensing Image

As remote sensing techniques developing, the demands of both higher resolution and data compression are generally growing. However, different from ordinary image, the remote sensing image not only contains large amounts of information which would results in compressing difficulties, but when under certain working environment, the compression algorithm should also have low complexity and power consumption. Until now, we are still far from the world's advanced level in the field of data compression. Therefore, the paper present here a lossy compression scheme for the remote sensing image with low bit rate.In the study, considering the specific characteristics of the remote sensing image, we proposed a lossy compression scheme for different type of images. The scheme using pre-process transform to decrease the influence of data distribution and improve the transform capability. We use DWT (Discrete Wavelet Transform) and utilize the lifting architecture to make it more effective and cost lower memory. The Rice coding and bit-plane coding are presented as the coding algorithm. This algorithm works well in low bit rate and real-time environment, while also meeting the requirement of low complexity and high quality at the same time.Additionally, for a lower bit rate and mass storage, to improve the observation feeling, we introduced the ROI coding scheme to alleviate the conflict between the compression ratio and quality. The algorithm developed the compression scheme, use the ROI-mask to transport the shape information and the code-blcok of ROI is coded at first. It is easy to implement and has nice ROI coding capability, also works well with the coding system proposed above.Overall, to solve the issue of ROI extracing, combined with the visual attention model, we present an automatic ROI extraction algorithm, which fit well with the original compression algorithm. It based on Itti's visual saliency model and is improved to suit the remote sensing image. It has satisfactory capability and is completely data-driven while no priori knowledge is required.