High-end PC monitors and TVs continue to increase their native display resolution to 4k by 2k and beyond. At the same time, high dynamic range formats demand higher bit depth for the underlying color component signals. Subsequently, uncompressed pixel amplitude processing becomes costly not only when transmitting over cable or wireless communication channels, but also across on-chip image processing pipelines that access external memory units. We recently presented a block-based non-linear memory compression architecture for text, graphics, and video enabling multi-dimensional error minimization with context sensitive control of visually noticeable artifacts. The underlying architecture was constrained to a small block size of 4x4 pixels. To increase compression ratio as well as image quality, we propose a novel approach that converts image amplitudes into a pair of discrete structure and magnitude quantities on a pixel by pixel basis which has been inspired by structure tensor analysis. Graceful degradation of image information is controlled by a single parameter which aims at optimally defining sparsity as a function of image context. Furthermore, we apply error diffusion via a threshold matrix to optimally diffuse the residual coding error. A detailed error distribution analysis and comparison with our previous algorithms highlights the effectiveness of our new approach, identifies its current limitations with regard to high quality color rendering, and illustrates algorithm specific visual artifacts.