Single Image Super-Resolution Based on Gradient Profile Sharpness

ABSTRACT

In this paper, a novel image superresolution algorithm is proposed based on GPS (Gradient Profile Sharpness). GPS is an edge sharpness metric, which is extracted from two gradient description models, i.e. a triangle model and a Gaussian mixture model for the description of different kinds of gradient profiles. Then the transformation relationship of GPSs in different image resolutions is studied statistically, and the parameter of the relationship is estimated automatically. Based on the estimated GPS transformation relationship, two gradient profile transformation models are proposed for two profile description models, which can keep profile shape and profile gradient magnitude sum consistent during profile transformation. Finally, the target gradient field of HR (high resolution) image is generated from the transformed gradient profiles, which is added as the image prior in HR image reconstruction model. Extensive experiments are conducted to evaluate the proposed algorithm in subjective visual effect, objective quality, and computation time. The experimental results demonstrate that the proposed approach can generate superior HR images with better visual quality, lower reconstruction error and acceptable computation efficiency as compared to state-of-the-art works 

Algorithm:

Super resolution  algorithm:

This Algorithm Used On Increasing Decreasing Resolution Purpose For Using.

HR:Higher Resolution Algorithm

Existing System                       

Single image super-resolution is a classic and active image processing problem, which aims to generate a high resolution image from a low resolution input image. Due to the severely under-determined nature of this problem, an effective image prior is necessary to make the problem solvable, and to improve the quality of generated images

Proposed System

  • More sophisticated interpolation models have also been proposed
  • To reduce the dependence on the training HR image, self-example based approaches were proposed, which utilized the observation that patches tended to redundantly recur inside an image within the same image scale as well as across different scales or there existed a transformation relationship across image space
  • . These approaches are more robust, however there are always some artifacts on their super-resolution results. Generally, the computational complexity of learning-based super-resolution approaches is quite high.
  • Various regularization terms have been proposed based on local gradient enhancement and globalgradient sparsity . Recently, metrics of edge sharpness have attracted researchers attention as the regularization term, since edges are of primary importance invisual image quality .
  • Based on the transformed GPS, two gradient profile transformation models are proposed, which can well keep profile shape and profile gradient magnitude sum consistent during the profile transformation.
  • Finally, the target gradient field of HR (high resolution) image is generated from transformed gradient profiles, which is added as the image priors in HR image reconstruction model.

MODULES

  • single image super-resolution
  • Gradient Profile Sharpness
  • Color Transfer
  • Multiple-reference color transfer
  • single image super-resolution:

Single-image super-resolution refers to the task of constructing a high-resolution enlargement of a given low-resolution image. Usual interpolation-based magnification introduces blurring. Then, the problem cast into estimating missing high-frequency details. Based on the framework of Freeman et al.

  1. interpolation of the input low-resolution image into the desired scale
  2. generation of a set of candidate images based on patch-wise regression: kernel ridge regression is utilized; To reduce the time complexity a sparse basis is found by combining kernel matching pursuit and gradient descent
  3. combining candidates to produce an image: patch-wise regression of output results in a set of candidates for each pixel location; An image output is obtained by combining the candidates based on estimated confidences for each pixel.
  4. post-processing based on the discontinuity prior of images: as a regularization method, kernel ridge regression tends to smooth major edges; The natural image prior proposed by Tappen et al. [2] is utilized to post-process the regression result such that the discontinuity at major edges are preserved.

Gradient Profile Sharpness:

A Novel edge sharpness metric GPS (gradient profile sharpness) is extracted as the eccentricity of gradient profile description models, which considers both the gradient magnitude and the spatial scattering of a gradient profile.

To precisely describe different kinds of gradient profile shapes, a triangle model and a mixed Gaussian model are proposed for short gradient profiles and heavy-tailed gradient profiles respectively. Then the pairs of GPS values under different image resolutions are studied statistically, and a linear GPS transformation relationship is formulated, whose parameter can be estimated automatically in each super-resolution application. Based on the transformed GPS, two gradient profile transformation models are proposed, which can well keep profile shape and profile gradient magnitude sum consistent during the profile transformation.

two gradient profile transformation models are proposed and the solve of HR image reconstruction model is introduced. Moreover, detailed experimental comparisons are made between the proposed approach and other state-of-the-art super-resolution methods, which are demonstrated in Section

Color Transfer:

Firstly proposed a way to match the means and variances between the target and the reference in the low correlated color space. This approach was efficient enough, but the simple means and variances  matching was likely to produce slight grain effect and serious color distortion. To prevent from the grain effect, Chang et al. proposed a color category based approach that categorized each pixelas one of the basic categories .Then a convex hull was generated in color space for each category of the pixel set, and the color transformation was applied with each pair of convex hull of the same category..

Multiple-reference color transfer:

requires the transfer naturally blending the colors from multiple references . However, as  illustrated  , the main difference exist among the references. Although both of the references are the sunshine theme, they have a big difference in the color appearance. This difference would easily lead to the grain effect in the result. As illustrated in , the  result has a serious grain effect approach adopts the gradient correction to suppress the grain, but it does not prevent the color distortion, see Our approach deals with the grain effect and distortion in each step, therefore, we can achieve a visual satisfactory result.

HARDWARE & SOFTWARE REQUIREMENTS:

HARDWARE REQUIREMENT:

v    Processor                                 –    Pentium –IV

  • Speed       –    1 GHz
  • RAM       –    256 MB (min)
  • Hard Disk      –   20 GB
  • Floppy Drive       –    44 MB
  • Key Board      –    Standard Windows Keyboard
  • Mouse       –    Two or Three Button Mouse
  • Monitor              –    SVGA

SOFTWARE REQUIREMENTS:

JAVA

  • Operating System        :           Windows XP or Win7
  • Front End       :           JAVA JDK 1.7
  • Back End :           MYSQL Server
  • Server :           Apache Tomact Server
  • Script :           JSP Script
  • Document :           MS-Office 2007

.NET

  • Operating System        :           Windows XP or Win7
  • Front End       :           Microsoft Visual Studio .NET 2008
  • Script :           C# Script
  • Back End :           MS-SQL Server 2005
  • Document :           MS-Office 2007