A combination of learning and non-learning based method for enhancement, compression and reconstruction of underwater images

作     者：Rashmi S.Nair Sandanam Domnic 

作者机构：Department of Computer ApplicationsNational Institute of TechnologyTiruchirappalliTamil Nadu620015India 

出 版 物：《Aquaculture and Fisheries》 (渔业学报（英文）)

年 卷 期：2022年第7卷第2期

页      面：201-210页

学科分类：090803[农学-渔业资源] 0908[农学-水产] 09[农学] 

基　　金：Centre of Excellence in Artificial Intelligence Kumoh National Institute of Technology, KIT, (Tiruchirappalli-15) 

主　　题：Convolutional neural network Discrete wavelet transform Residual dense convolutional neural network Peak signal to noise ratio Structural similarity index metric Super resolution CNN 

摘      要：Complex nature of underwater environment poses biggest challenge towards image acquisition and transmission of underwater images.This paper proposes an integrated approach which consists of a non-learning enhancement method with deep Convolutional Neural Networks(CNN)for compression and reconstruction of the image.The proposed method does color and contrast correction for image enhancement.The enhanced images are down-sampled using 9-layer CNN followed by Discrete Wavelet Transform(DWT).The decompression is done by using Inverse DWT.Further,the sub-pixel up-sampled image is de-blurred using a three-layer CNN.Residual Dense CNN(RD-CNN)is used to improve the quality of the reconstructed image after deblurring.The quality of the reconstructed images is measured using Peak Signal to Noise Ratio(PSNR)and Structural Similarity Index Metric(SSIM).The proposed model provides better image enhancement,compression,and reconstruction quality than the existing state-of-the-art methods and Super Resolution CNN(SRCNN)respectively.