Gradient Magnitude Based Normalised Convolution

Although image data can often be sparse for a variety of different reasons, standard image processing techniques require the use of complete image data. Therefore sparse image data must undergo reconstruction to yield complete images prior to any subsequent processing. Highly accurate image reconstruction techniques tend to be expensive to implement whilst simpler techniques, such as image interpolation, are usually not adequate to support subsequent reliable image processing. A common approach to image reconstruction is normalised convolution; we present a modified approach to normalised convolution which is based on the sparse image content and we demonstrate that accurate reconstruction is achieved yielding better image processing results than the current standard normalised convolution.