A Brief Review on Recent Trends in Image Restoration

A Brief Review on Recent Trends in Image Restoration

Saurav Prakash (National Institute of Technology, India)
Copyright: © 2016 |Pages: 16
DOI: 10.4018/978-1-4666-8789-9.ch008
OnDemand PDF Download:
Available
$37.50
Current Special Offers
No Current Special Offers
TOTAL SAVINGS: $37.50

Abstract

This chapter gives the opportunity to get an idea of recent trends in image denoising and restoration. It relates to the present research scenario in the field of image restoration. As much as possible the newest break-through regarding the methods of denoising as well as the performance metrics of evaluation has been dealt. The assessments done by the researchers have been included first so as to know how much analysis they propose to be done with respect to the application point of view of the denoising methods. The concept behind the metric selection for the assessment and evaluation has been introduced along with the need for shifting the dependence of the research community towards the newly proposed metrics than the old ones. The new trends in image denoising have been referred duly so that the readers can directly refer to the main algorithms and techniques from the papers proposed by their authors.
Chapter Preview
Top

Choice Of The Performance Metrics

The choice of the visual fidelity metric for a particular case varies from the test algorithm, the complexity of calculation and subjective judgments made by humans’ perceptions. The Performance metrics have a literature of their own, According to (Chandler & Hemami, 2007) fundamentally, the metric of evaluation has been developed either on bottom-up properties of vision or by relying on how our visual system responds to a distorted image. These metric of evaluations can be divided as:

  • 1.

    Mathematically Convenient Metrics: These take in to account the intensity of distortions; for example: Mean-Squared Error (MSE), Signal to Noise Ratio (SNR), Peak Signal-to-Noise Ratio (PSNR), Root Mean-Squared Error (RMSE), and so forth.

  • 2.

    Metrics Based on Near-Threshold Psychophysics: These take in to account a frequency-decomposition algorithm, which uses the contrast detection thresholds along with the elevations in the thresholds due to masking effects imposed by images. For example, weighted MSE or activity based measures (Teo & Heeger, 1994; Lai & Kuo, 1997; Winkler, 1999).

  • 3.

    Metrics Based on Overarching Principles: Overreaching principles could be structural or information extraction. The basic principle is that if structural content (such as object boundaries or regions of high entropy etc.) most closely matches that of the original image, the image could be considered a high quality image. (Wang, Bovik, Sheikh, & Simoncelli, 2004; Sheikh, Bovik, & Veciana, 2005; Zhai, Zhang, Yang, & Xu, 2005; Shnayderman, Gusev, & Eskicioglu, 2006).

Complete Chapter List

Search this Book:
Reset