Image Super Resolution Based On Contextual Sparse Representation

As the rapid development of Internet communication technology, the expectationtowards information is getting higher. Compared with original communication methodbased on text and sound, multimedia communication combined multiplied informationsource of sound, images, videos and etc., has incomparable advantage. It becomes moreand more important in the field of daily communication and it definitely will be theprime path for information technology. The image deterioration due to acquisitiondevice or transition is a key problem to solve to increase the quality of multimediacommunication, and image super resolution provides a good solution.Sparse representation based on compressive sensing theory has good performancein many fields of digital pictures processing. With the uniqueness of sparserepresentation, image super resolution is well solved. Based on the traditional imagesuper resolution method with sparse representation, a sparse representation methodbased on context sparse representation is raised in this paper to make a better use ofcontext in the application of image super resolution frame, so that the performance ofsparse representation is enhanced and finally a reconstructed image with better quality isacquainted. This paper mainly covers the following research:First, a precise mathematical model aiming to traditional image super resolutionmethod based on sparse representation is built, and the reconstruction-restricted imageenhancement is added to image recovering frame based on sparse prior. Besides, afteranalyzing traditional methods, a strategy that adds context information to original imagereconstructed frame is forwarded to increase the performance of sparse representation.Then, the image super resolution method based on multi-dictionary sparserepresentation is brought forward in this paper to realize integrate context informationinto traditional image super resolution frame based on sparse representation. This avoidsthe disadvantage of traditional method that dictionary has strong representationcapability towards large quantity of data, while acts poorly when the quantity is low.At last, the image super resolution method based on multi-dictionary sparserepresentation is further improved and training model for discriminative dictionary israised. Considering dictionary redundancy and cost of multi-dictionaries based superresolution method, the multi-dictionary frame is replaced by dictionary through trainingwith Fisher judging rule to integrate context information into image super sparserepresentation frame. Fisher judging rule can effectively increase the discriminatingcapability, keeping every data type has a set of discriminative bases.