UNDERWATER IMAGE ENHANCEMENT AND DEHAZING FOR SUBSEA PIPELINE CORROSION INSPECTION

The subsea pipeline corrosion in offshore oil and gas industry is considered as an important problem that further leads to cracks and leakages. Often, the human divers are involved to perform the visual inspection tasks for pipeline integrity management on regular basis. However, the offshore setups are moving from shallow water to deep water due to exhaustion of oil and gas reservoirs. In deep water, the inspection by human divers is inviable due to inhospitable environmental conditions. Due to the reliability of visual inspection method, image based-solutions are found to be effective for deep water inspection activities. However, the underwater image is degraded by low contrast, color attenuation and blurring issues by suspended particles in water. Hence the underwater image enhancement and dehazing operations are required prior to establishing image based-inspection. This research is focused on visual inspection for corrosion estimation using image data. Therefore, this thesis proposes underwater image restoration, with contributions in color and contrast enhancement as well as dehazing for a final application that targets the estimation of the corroded surface of the pipeline. The blur is estimated using a developed methodology through a neural network on simulated data. Furthermore, a Wiener filter is used through deconvolution process by considering the estimated blur. Once dehazed, the image is further enhanced in the wavelet domain. Finally, the region of interest is clustered into three clusters to estimate different degree of corrosion. The method has been tested on real images with simulated blur and validated both visually and quantitatively with the existing method using publicly available underwater images. The validated results proved the performance of the proposed method in terms of SSIM and MoE as compared to existing solution. The improvement in SSIM and MoE is calculated as 57% and 6.7% respectively. The accuracy of corrosion estimation in all three clusters is above 90% which suggested to implement the proposed method for real-time corrosion inspection application in oil and gas industry.