Underwater Image Enhancement for Instance Segmentation using Deep Learning Models

Abstract

Underwater instance segmentation greatly depends on color-blended underwater images. In this work, a combination of Generalized Color Fourier Descriptor (GCFD), Convolutional Neural Network (CNN) and Mask Region-based Convolutional Neural Network (Mask R-CNN) models were employed to generate a mask for each bounding-boxed Region of Interest (ROI) to obtain enhanced individual underwater segmented images from their complex background accurately. By this image enhancement approach, individual underwater instances are segmented from their complex background accurately. The Patch-based Contrast Quality Index (PCQI) evaluation of our proposed image enhancement method (GCFD) after conducting experiment on the employed datasets shows performance accuracy of 1.1336, which is higher than the 1.1126 performance accuracy achieved by the Contrast-enhancement Algorithm (CA).

Fine-Grained Analysis of Coral Instance Segmentation using YOLOv8 Models

“…Challenges arising from the complex and often color-distorted nature of underwater imagery have also been addressed such as the work of (Bello et al, 2023) who proposed a novel approach combining Generalized Color Fourier Descriptor (GCFD), Convolutional Neural Networks (CNN), and Mask R-CNN to improve segmentation accuracy in underwater environments. In addition to coral reefs, (Lv et al, 2022) also applied neural networks to improve the detection of small and camouflaged marine organisms, demonstrating the versatility of this technique for various underwater animal analysis tasks.…”

“…YOLOv8 can achieve efficient segmentation, due to its ideal encoder-decoder architecture to capture global context information in images (Yang et al, 2023). This capability overcomes the limitations of existing methods, which often have difficulty accurately identifying individual corals on densely populated or geometrically complex reefs (Bello et al, 2023).…”