Contact at mumbai.academics@gmail.com or 8097636691
Responsive Ads Here

Wednesday, 7 February 2018

Single Image Superresolution Based on Gradient Profile Sharpness

Single Image Superresolution Based on Gradient Profile Sharpness

Abstract—Single image super resolution is a classic and active image processing problem, which aims to generate a high-resolution (HR) 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. Single Image Superresolution Based on Gradient Profile project, a novel image super resolution algorithm is proposed based on gradient profile sharpness (GPS).
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 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 with state-of-the-art works. Single Image Super resolution Based on Gradient Profile Sharpness
CONCLUSION
Single Image Super resolution Based on Gradient Profile Sharpness project, a novel single image super-resolution algorithm is proposed based on the edge sharpness metric of GPS. Two gradient profile description models are proposed for representing gradient profiles with different lengths and different complicated shapes. Then, GPS is defined, the GPS transformation relationship is studied statistically, and a method is proposed to estimate the parameter of GPS transformation relationship automatically. Based on the transformed GPS, two gradient profiles transformation models are proposed, which can keep the profile magnitude sum and profile shape consistent during the transformation. Finally, the transformed gradients are utilized as priors in the high resolution image reconstruction. Plenty of experiments are conducted to evaluate the performance of the proposed method on subjective visual quality, objective quality, and computation time. Experimental results show that the proposed approach can faithfully recover high-resolution image with little observable artifacts.

No comments:

Post a Comment