A hybrid underwater crack image enhancement method.

This study proposes a hybrid underwater crack image processing method. The first half of the algorithm is based on traditional image processing, which analyzes and judges the color of an image, establishes the image chromaticity factor K, and performs color correction when K is greater than a threshold value. A quadtree analysis method is used to estimate the background light of the image. The latter half of the algorithm is based on deep learning (DL), using convolutional neural networks (CNNs) to learn image features, adopting an edge feature extraction network to extract edge feature information based on the characteristics of underwater crack images, and finally the image features are fused with the edge feature. Depthwise separable convolution and pixel attention mechanisms were used in CNNs to improve feature extraction ability while reducing computational complexity. Ultimately, crack image restoration relies on the underwater image model to achieve enhanced clarity. This hybrid algorithm combined the interpretability of traditional algorithms with the generality of DL algorithms. Compared with various traditional image enhancement and DL algorithms, the proposed algorithm achieved good performance in terms of parameters such as peak signal-to-noise ratio, structural similarity, underwater image quality measure and pixel-based contrast quality Index.