Image Dehazing Using Machine Learning Methods

  • 王 峻國

Student thesis: Doctoral Thesis

Abstract

In recent years the image dehazing issue has been widely discussed During photography in an outdoor environment the medium in the air causes light attenuation and reduce image quality; these impacts are especially obvious in a hazy environment Reduction of image quality results in the loss of information which hinders image recognition systems to identify objects in the image Removal of haze can provide a reference for subsequent image processing for specific requirements Notably image dehazing technology is used to maintain image quality during preprocessing This dissertation presents machine learning methods for image haze removal and consists of two major parts In the first part a fuzzy inference system (FIS) model is presented Users of this model can customize designs to generate applicable fuzzy rules from expert knowledge or data The number of fuzzy rules is fixed In addition the FIS model requires substantial amounts of data and expertise; even if the model is used to develop a fuzzy system the image output of that system may suffer from a loss of accuracy Therefore in the second part of this dissertation a recurrent fuzzy cerebellar model articulation controller (RFCMAC) model with a self-evolving structure and online learning is presented to improve the FIS model The recurrent structure in an RFCMAC is formed with internal loops and internal feedback by feeding the rule firing strength of each rule to other rules and to itself A Takagi-Sugeno-Kang (TSK) type is used in the consequent part of the RFCMAC The online learning algorithm consists of structure and parameter learning The structure learning depends on an entropy measure to determine the number of fuzzy rules The parameter learning based on back-propagation can adjust the shape of the membership function and the corresponding weights of the consequent part This dissertation describes the proposed machine learning methods and its related algorithm applies them to various image dehazing problems and analyzes the results to demonstrate the effectiveness of the proposed methods
Date of Award2016 Aug 22
Original languageEnglish
SupervisorShen-Chuan Tai (Supervisor)

Cite this

'