Extremely Low and Variable Bandwidth Image Compression with Region of Interest Applied to Real Time Underwater Robotic Interventions
Eduardo Moscoso Rubino · 2018
A new fast and progressive set-partitioning image compression parallel algorithm with Region Of Interest (ROI) which outputs an embedded bit oriented rate-distortion optimized stream and addresses very low bit rate compression is presented.User defined variable packet sizes make it suitable for the implementation of any communications protocols, either underwater or in any other scenario, while remaining competitive with current state-of-the-art compressors at higher bit rates.A parallel algorithm for the Discrete Wavelet Transform (DWT) based on the lifting scheme is also presented and it is shown to be optimal in the sense that no other implementation may be faster if memory saturation is achieved.The best ordering for the significant and refinement bits of the transform coefficients is derived, using the Mean Squared Error (MSE) as the error measure, by fitting a Probability Density Function (PDF) to the transform coefficients and weighting the error for each range of coefficients by its respective DWT subband gain.A general scheme for Region Of Interest (ROI), including a non-linear scaling ROI, is presented in which the lower bitplanes of the foreground coefficients are delayed in exchange for better background reconstruction, achieving a more effective blending of foreground and background information.Finally, an implementation for both 32-bit and 64-bit ARM and x86 architectures was validated in an actual wireless underwater robotic teleoperation context.