Non-local enhancement of electronic portal imaging with tensor texture descriptors

Electronic portal imaging for reliable patient setup verification is a very important task in radiation therapy. In-treatment imaging with guaranteed visual quality is crucial to track relevant anatomy features. However, on-line images are often blurry because of imaging physics. In this reported work, a robust multi-scale approach for image enhancement is developed. An edge-aware function is investigated with a tensor of a Gaussian pyramid and measured non-locally as the output coefficient of a Laplacian pyramid, which controls how the edges are modified. Furthermore, a greyscale stretch is performed in the coarsest level of the Gaussian pyramid to produce guaranteed holistic contrast. Experimental results demonstrate the stability and simplicity of the proposed method in improving visual quality and preserving important textures.