Single Image Super-Resolution Based on Gradientprofile Sharpness

Single Image Super-Resolution Based on GradientProfile Sharpness

Abstract

Single image super-resolution is a classic and activeimage processing problem, which aims to generate a high resolutionimage from a low resolution input image. Due to the severelyunder-determined nature of this problem, an effective image prioris necessary to make the problem solvable, and to improve thequality of generated images. In this paper, a novel image superresolutionalgorithm is proposed based on GPS (Gradient ProfileSharpness). GPS is an edge sharpness metric, which is extractedfrom two gradient description models, i.e. a triangle model and aGaussian mixture model for the description of different kinds ofgradient profiles. Then the transformation relationship of GPSsin different image resolutions is studied statistically, and the

parameter of the relationship is estimated automatically. Basedon the estimated GPS transformation relationship, two gradientprofile transformation models are proposed for two profiledescription models, which can keep profile shape and profilegradient magnitude sum consistent during profile transformation.Finally, the target gradient field of HR (high resolution) image isgenerated from the transformed gradient profiles, which is addedas the image prior in HR image reconstruction model. Extensiveexperiments are conducted to evaluate the proposed algorithmin subjective visual effect, objective quality, and computationtime. The experimental results demonstrate that the proposedapproach can generate superior HR images with better visual

quality, lower reconstruction error and acceptable computationefficiency as compared to state-of-the-art works.

Existing Method:

To reduce the dependenceon the training HR image, self-example based approaches were

proposed, which utilized the observation that patches tended toredundantly recur inside an image within the same image scaleas well as across different scales

Demerits

High complexity

Proposed Method

novel image superresolutionalgorithm is proposed based on GPS (Gradient ProfileSharpness). GPS is an edge sharpness metric, which is extractedfrom two gradient description models, i.e. a triangle model and aGaussian mixture model for the description of different kinds ofgradient profiles.

Merits:

Execution time is less

Further Details Contact: A Vinay 9030333433, 08772261612

Email: |