Light Weight Encoder-Decoder for Underwater Images in Internet of Underwater Things (IoUT)

Internet of Underwater Things (IoUT) explores the applications of Internet of Things (IoT) to monitor the sea animal habitat, observe atmosphere, and predict defense and predict defense and disaster. Raw underwater images are affected by absorption and dispersal of light due to underwater environment. Low power computational devices are preferred to cut down the cost of IoUT devices. Because of underwater environment nature, transmission of underwater images captured by underwater devices is considered as a big challenge. There is a need to provide solutions to amplify color, contrast and brightness aspects of captured underwater images to provide good visual understanding. Conventional compression techniques used for terrestrial environment, causes ringing artefacts due to the variable characteristics of underwater images. Deep image compression techniques consume more computational power and time, making them least efficient for low power computational devices. In this study, a low computational power and less time-consuming image compression technique is proposed to achieve high encoding efficiency and good reconstruction quality of underwater images. The proposed technique suggests using Convolutional Neural Network (CNN) at encoder side, which compresses and retains the structural data of the underwater image. And relative global histogram stretching based technique has been used at the decoder side to enhance the reconstructed underwater image. The proposed methodology is compared with conventional methods like Joint Pictures Experts Group (JPEG), Better Portable Graphics (BPG), Contrast Limited Adaptive Histogram Equalization (CLAHE) and deep learning techniques like Super Resolution Convolutional Neural Network (SRCNN) and Residual encoder-decoder methods to evaluate the reconstructed image quality. The presented work provides high quality image in comparison with both conventional and SRCNN method.