Faithful visualization of categorical data

Data is today collected in very large amounts from various kind of sources: shopping, genetic tests, etc. Given the amount of data collected, one might be interested in visualizing that data in a meaningful way. Due to their size, producing an image of such datasets is not easy as the image has to fit on the screen. Compression of the image is needed in order to have fewer elements to display. However, the quality of this compression greatly depends on the structure of the data. Given a binary matrix represented by white and black pixels, well separated groups of pixels (e.g. black and white points are respectively gathered together) gives a lower loss of information because summarizing that data concisely is easier. Moreover, reorganizing the rows and columns of a dataset can give interesting insights in structures hidden within the data. In this thesis, we introduce an innovative algorithm to reorder binary and categorical matrices using convolution. We then use the results of the algorithm to display the datasets faithfully.