Tensorial Orientation Scores

Orientation scores [7, 10] are representations of images built using filters that only select on orientation (and not on the magnitude of the frequency). Importantly, they allow (easy) reconstruction, making them ideal for use in a filtering pipeline. Traditionally a specific set of orientations has to be chosen, and the response is determined for those orientations. This work introduces an alternative, where a tensorial representation is built that approximates an idealized orientation score in a well-defined way. It is shown that the filter’s output can be usefully interpreted in terms of tensor decompositions. Tensorial orientation scores can be considered to fit in a family of filtering schemes that includes not just traditional orientation scores, but also monomial filters [16] and curvelets [3].

• Type: Journal
• Year: 2015
• Where: CAIP
• Authors: Jasper J. van de Gronde
• Keywords: CAIP 2015, Image Analysis