A Novel Adaptive Compression System for LCS Unmanned Vehicle Sensor Data Compression

Most sonar and camera images in practice are inherently sparse yet highly non-stationary. Optimal prediction, sampling, representation, and estimation of such signals require locally adaptive decomposition that can quickly capture the signal characteristics within a small neighborhood. In order to achieve the demanding required level of compression performance, we propose an adaptive block-based coding framework the strategy of choice of the most recent international image compression standard JPEG-XR as well as the intra-frame coding method in current international video compression standards H.264/H.265. In a nutshell, we propose to combine the best features of these well-established state-of-the-art coders and then add on top an unprecedented level of flexibility and adaptivity on a block-by-block basis. We will investigate a data-adaptive context-adaptive approach where each operator in the block-coding framework is optimized and selected based on the local statistics/characteristics of the given input image. The encoder will take into account the bit budget requirement for each block while minimizing its reconstructed mean-square error. Our compression is also robust and error resilient. It should be emphasized that our proposed framework is applicable to both lossless and lossy compression.