An Enhanced Recurrent Neural Network for Image Deraining

Abstract

Computer vision algorithms like object detection image classification have changed our life significantly in various applications However these algorithms cannot usually have good performance in practical applications due to the fact that the unpredictable degradations often occur in realistic scene for instance noise illumination and severe weather conditions Commonly seen weather conditions such as rain snow and sandstorm can adversely affect the performance of many computer vision tasks In this Thesis a rain removal algorithm based on recurrent neural network is proposed to remove rain streak stage by stage A Rain Streak Prediction Network is proposed to predict the rain streak part of a rainy image providing more information to deraining A residual dense block combining with channel attention mechanism called RDCAB is used to enhance the ability of deraining Experimental results show that the proposed method gets more nature textures and details compared with available methods

Date of Award	2020
Original language	English
Supervisor	Shen-Chuan Tai (Supervisor)